Hyperspectral Remote Sensing of Vegetation, 4-Volume Set [2 ed.] 1138066257, 9781138066250

Table of contents :
Cover
Volume1
Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Table of Contents
Foreword to the First Edition
Foreword to the Second Edition
Preface
Acknowledgments
Editors
Contributors
Acronyms and Abbreviations
Section I: Introduction to Hyperspectral Remote Sensing of Agricultural Crops and Vegetation
Chapter 1: Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops
Section II: Hyperspectral Sensor Systems
Chapter 2: Hyperspectral Sensor Characteristics: Airborne, Spaceborne, Hand-Held, and Truck-Mounted; Integration of Hyperspectral Data with LiDAR
Chapter 3: Hyperspectral Remote Sensing in Global Change Studies
Section III: Hyperspectral Libraries of Agricultural Crops and Vegetation
Chapter 4: Monitoring Vegetation Diversity and Health through Spectral Traits and Trait Variations Based on Hyperspectral Remote Sensing
Chapter 5: The Use of Hyperspectral Proximal Sensing for Phenotyping of Plant Breeding Trials
Chapter 6: Linking Online Spectral Libraries with Hyperspectral Test Data through Library Building Tools and Code
Chapter 7: The Use of Spectral Databases for Remote Sensing of Agricultural Crops
Chapter 8: Characterization of Soil Properties Using Reflectance Spectroscopy
Section IV: Hyperspectral Data Mining, Data Fusion, and Algorithms
Chapter 9: Spaceborne Hyperspectral EO-1 Hyperion Data Pre-Processing: Methods, Approaches, and Algorithms
Chapter 10: Hyperspectral Image Data Mining
Chapter 11: Hyperspectral Data Processing Algorithms
Chapter 12: Methods for Linking Drone and Field Hyperspectral Data to Satellite Data
Chapter 13: Integrating Hyperspectral and LiDAR Data in the Study of Vegetation
Chapter 14: Fifty-Years of Advances in Hyperspectral Remote Sensing of Agriculture and Vegetation—Summary, Insights, and Highlights of Volume I: Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation
Index
Volume2
Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
of Contents
Foreword to the First Edition
Foreword to the Second Edition
Preface
Acknowledgments
Editors
Contributors
Acronyms and Abbreviations
Section I: Hyperspectral Vegetation Indices
Chapter 1: Hyperspectral Vegetation Indices
Chapter 2: Derivative Hyperspectral Vegetation Indices in Characterizing Forest Biophysical and Biochemical Quantities
Section II: Hyperspectral Image Classification Methods and Approaches
Chapter 3: Advances in Hyperspectral Image Classification Methods for Vegetation and Agricultural Cropland Studies
Chapter 4: Big Data Processing on Cloud Computing Architectures for Hyperspectral Remote Sensing
Section III: Hyperspectral Vegetation Index Applications to Agriculture and Vegetation
Chapter 5: Noninvasive Quantification of Foliar Pigments
Chapter 6: Hyperspectral Remote Sensing of Leaf Nitrogen Concentration in Cereal Crops
Chapter 7: Optical Remote Sensing of Vegetation Water Content
Chapter 8: Estimation of Nitrogen Content in Herbaceous Plants Using Hyperspectral Vegetation Indices
Chapter 9: Hyperspectral Remote Sensing of Leaf Chlorophyll Content
Section IV: Conclusions
Chapter 10: Fifty Years of Advances in Hyperspectral Remote Sensing of Agriculture and Vegetation—Summary, Insights, and Highlights of Volume II
Index
Volume3
Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Table of Contents
Foreword to the First Edition
Foreword to the Second Edition
Preface
Acknowledgments
Editors
Contributors
Acronyms and Abbreviations
Section I: Vegetation Biophysical and Biochemical Properties
Chapter 1: Recent Developments in Remote Estimation of Crop Biophysical and Biochemical Properties at Various Scales
Chapter 2: Hyperspectral Assessment of Ecophysiological Functioning for Diagnostics of Crops and Vegetation
Chapter 3: Spectral and Spatial Methods for Hyperspectral and Thermal Image-Analysis to Estimate Biophysical and Biochemical Properties of Agricultural Crops
Chapter 4: Spectral and 3D Nonspectral Approaches to Crop Trait Estimation Using Ground and UAV Sensing
Chapter 5: Photosynthetic Efficiency and Vegetation Stress
Section II: Plant Species Identification and Discrimination
Chapter 6: Crop Type Discrimination Using Hyperspectral Data: Advances and Perspectives
Chapter 7: Identification of Canopy Species in Tropical Forests Using Hyperspectral Data
Chapter 8: Characteristics of Tropical Tree Species in Hyperspectral and Multispectral Data
Chapter 9: Detecting and Mapping Invasive Plant Species Using Hyperspectral Data
Chapter 10: Visible, Near Infrared, and Thermal Spectral Radiance On-Board UAVs for High-Throughput Phenotyping of Plant Breeding Trials
Section III: Conclusions
Chapter 11: Fifty Years of Advances in Hyperspectral Remote Sensing of Agriculture and Vegetation—Summary, Insights, and Highlights of Volume III Biophysical and Biochemical Characterization and Plant Species Studies
Index
Volume4
Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Table of Contents
Foreword to the First Edition
Foreword to the Second Edition
Preface
Acknowledgments
Editors
Contributors
Acronyms and Abbreviations
Section I: Detecting Crop Management Practices, Plant Stress, and Disease
Chapter 1: Using Hyperspectral Data in Precision Farming Applications
Chapter 2: Hyperspectral Narrowbands and Their Indices in Study of Nitrogen Content of Cotton Crops
Chapter 3: Analysis of the Effects of Heavy Metals on Vegetation Hyperspectral Reflectance Properties
Section II: Vegetation Processes and Function (ET, Water Use, GPP, LUE, Phenology)
Chapter 4: Mapping the Distribution and Abundance of Flowering Plants Using Hyperspectral Sensing
Chapter 5: Crop Water Productivity Estimation with Hyperspectral Remote Sensing
Chapter 6: Hyperspectral Remote Sensing Tools for Quantifying Plant Litter and Invasive Species in Arid Ecosystems
Chapter 7: Hyperspectral Applications to Landscape Phenology
Section III: Land Cover, Forests, and Wetland and Urban Applications Using Hyperspectral Data
Chapter 8: The Use of Hyperspectral Earth Observation Data for Land Use/Cover Classification: Present Status, Challenges, and Future Outlook
Chapter 9: Hyperspectral Remote Sensing for Forest Management
Chapter 10: Characterization of Pastures Using Field and Imaging Spectrometers
Chapter 11: Hyperspectral Remote Sensing of Wetland Vegetation
Section IV: Thermal, SWIR, and Visible Remote Sensing
Chapter 12: Hyperspectral Remote Sensing of Fire: A Review
Section V: Hyperspectral Data in Global Change Studies
Chapter 13: Hyperspectral Data in Long-Term, Cross-Sensor Continuity Studies
Section VI: Hyperspectral Remote Sensing of Other Planets
Chapter 14: Hyperspectral Analysis of Rocky Surfaces on Earth and Other Planetary Bodies
Section VII: Conclusions
Chapter 15: Fifty Years of Advances in Hyperspectral Remote Sensing of Agriculture and Vegetation—Summary, Insights, and Highlights of Volume IV: Advanced Applications in Remote Sensing of Agricultural Crops and Natural Vegetation
Index

Citation preview

HYPERSPECTRAL REMOTE SENSING OF VEGETATION SECOND EDITION VOLUME I

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Hyperspectral Remote Sensing of Vegetation Second Edition

Volume I: Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation Volume II: Hyperspectral Indices and Image Classifications for Agriculture and Vegetation Volume III: Biophysical and Biochemical Characterization and Plant Species Studies Volume IV: Advanced Applications in Remote Sensing of Agricultural Crops and Natural Vegetation

HYPERSPECTRAL REMOTE SENSING OF VEGETATION SECOND EDITION VOLUME I

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Edited by

Prasad S. Thenkabail John G. Lyon Alfredo Huete

CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2019 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group, an Informa business No claim to original U.S. Government works Printed on acid-free paper International Standard Book Number-13: 978-1-138-05854-5 (Hardback) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged, please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http:// www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com

Dr. Prasad S. Thenkabail, Editor-in-Chief of these four volumes would like to dedicate the four volumes to three of his professors at the Ohio State University during his PhD days: 1. Late Prof. Andrew D. Ward, former professor of The Department of Food, Agricultural, and Biological Engineering (FABE) at The Ohio State University, 2. Prof. John G. Lyon, former professor of the Department of Civil, Environmental and Geodetic Engineering at the Ohio State University, and 3. Late Prof. Carolyn Merry, former Professor Emerita and former Chair of the Department of Civil, Environmental and Geodetic Engineering at the Ohio State University.

Contents Foreword to the First Edition ............................................................................................................ix Foreword to the Second Edition..................................................................................................... xiii Preface............................................................................................................................................xvii Acknowledgments ...........................................................................................................................xxi Editors ............................................................................................................................................xxv Contributors ..................................................................................................................................xxix Acronyms and Abbreviations..................................................................................................... xxxiii

Section i introduction to Hyperspectral Remote Sensing of Agricultural crops and Vegetation Chapter 1

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops........................................................................................................3 Prasad S. Thenkabail, John G. Lyon, and Alfredo Huete

Section ii Hyperspectral Sensor Systems Chapter 2

Hyperspectral Sensor Characteristics: Airborne, Spaceborne, Hand-Held, and Truck-Mounted; Integration of Hyperspectral Data with LiDAR ............................. 41 Fred Ortenberg

Chapter 3

Hyperspectral Remote Sensing in Global Change Studies ........................................ 71 Jiaguo Qi, Yoshio Inoue, and Narumon Wiangwang

Section iii Hyperspectral Libraries of Agricultural crops and Vegetation Chapter 4

Monitoring Vegetation Diversity and Health through Spectral Traits and Trait Variations Based on Hyperspectral Remote Sensing ................................................. 95 Angela Lausch and Pedro J. Leitão

Chapter 5

The Use of Hyperspectral Proximal Sensing for Phenotyping of Plant Breeding Trials ......................................................................................................... 127 Andries B. Potgieter, James Watson, Barbara George-Jaeggli, Gregory McLean, Mark Eldridge, Scott C. Chapman, Kenneth Laws, Jack Christopher, Karine Chenu, Andrew Borrell, Graeme L. Hammer, and David R. Jordan

vii

viii

Chapter 6

Contents

Linking Online Spectral Libraries with Hyperspectral Test Data through Library Building Tools and Code............................................................................. 149 Muhammad Al-Amin Hoque and Stuart Phinn

Chapter 7

The Use of Spectral Databases for Remote Sensing of Agricultural Crops ............ 165 Andreas Hueni, Lola Suarez, Laurie A. Chisholm, and Alex Held

Chapter 8

Characterization of Soil Properties Using Reflectance Spectroscopy ..................... 187 E. Ben-Dor, S. Chabrillat, and José A. M. Demattê

Section iV Hyperspectral Data Mining, Data Fusion, and Algorithms Chapter 9

Spaceborne Hyperspectral EO-1 Hyperion Data Pre-Processing: Methods, Approaches, and Algorithms.................................................................................... 251 Itiya P. Aneece, Prasad S. Thenkabail, John G. Lyon, Alfredo Huete, and Terrance Slonecker

Chapter 10 Hyperspectral Image Data Mining........................................................................... 273 Sreekala G. Bajwa, Yu Zhang, and Alimohammad Shirzadifar Chapter 11 Hyperspectral Data Processing Algorithms ............................................................. 303 Antonio Plaza, Javier Plaza, Gabriel Martín, and Sergio Sánchez Chapter 12 Methods for Linking Drone and Field Hyperspectral Data to Satellite Data .......... 321 Muhammad Al-Amin Hoque and Stuart Phinn Chapter 13 Integrating Hyperspectral and LiDAR Data in the Study of Vegetation ................. 355 Jessica J. Mitchell, Nancy F. Glenn, Kyla M. Dahlin, Nayani Ilangakoon, Hamid Dashti, and Megan C. Maloney Chapter 14 Fifty-Years of Advances in Hyperspectral Remote Sensing of Agriculture and Vegetation—Summary, Insights, and Highlights of Volume I: Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation ..................... 395 Prasad S. Thenkabail, John G. Lyon, and Alfredo Huete Index .............................................................................................................................................. 439

Foreword to the First Edition The publication of this book, Hyperspectral Remote Sensing of Vegetation, marks a milestone in the application of imaging spectrometry to study 70% of the Earth’s landmass which is vegetated. This book shows not only the breadth of international involvement in the use of hyperspectral data but also in the breadth of innovative application of mathematical techniques to extract information from the image data. Imaging spectrometry evolved from the combination of insights from the vast heterogeneity of reflectance signatures from the Earth’s surface seen in the ERTS-1 (Landsat-1) 4-band images and the field spectra that were acquired to help fully understand the causes of the signatures. It wasn’t until 1979 when the first hybrid area-array detectors, mercury-cadmium-telluride on silicon CCD’s, became available that it was possible to build an imaging spectrometer capable of operating at wavelengths beyond 1.0 µm. The AIS (airborne imaging spectrometer), developed at NASA/JPL, had only 32 cross-track pixels but that was enough for the geologists clamoring for this development to see between the bushes to determine the mineralogy of the substrate. In those early years, vegetation cover was just a nuisance! In the early 1980s, spectroscopic analysis was driven by the interest to identify mineralogical composition by exploiting absorptions found in the SWIR region from overtone and combination bands of fundamental vibrations found in the mid-IR region beyond 3 µm and the electronic transitions in transition elements appearing, primarily, short of 1.0 µm. The interests of the geologists had been incorporated in the Landsat TM sensor in the form of the add-on, band 7 in the 2.2 µm region based on field spectroscopic measurements. However, one band, even in combination with the other six, did not meet the needs for mineral identification. A summary of mineralogical analyses is presented by Vaughan et al. in this volume. A summary of the historical development of hyperspectral imaging can be found in Goetz (2009). At the time of the first major publication of the AIS results (Goetz et al., 1985), very little work on vegetation analysis using imaging spectroscopy had been undertaken. The primary interest was in identifying the relationship of the chlorophyll absorption red-edge to stress and substrate composition that had been seen in airborne profiling and in field spectral reflectance measurements. Most of the published literature concerned analyzing NDVI, which only required two spectral bands. In the time leading up to the 1985 publication, we had only an inkling of the potential information content in the hundreds of contiguous spectral bands that would be available to us with the advent of AVIRIS (airborne visible and infrared imaging spectrometer). One of the authors, Jerry Solomon, presciently added the term “hyperspectral” to the text of the paper to describe the “…multidimensional character of the spectral data set,” or, in other words, the mathematically, over-determined nature of hyperspectral data sets. The term hyperspectral as opposed to multispectral data moved into the remote sensing vernacular and was additionally popularized by the military and intelligence community. In the early 1990s, as higher quality AVIRIS data became available, and the first analyses of vegetation using statistical techniques borrowed from chemometrics, also known as NIRS analysis used in the food and grain industry, were undertaken by John Aber and Mary Martin of the University of New Hampshire. Here, nitrogen contents of tree canopies were predicted from reflectance spectra by regression techniques using reference measurements from laboratory wet chemical analyses of needle and leaf samples acquired by shooting down branches. At the same time, the remote sensing community began to recognize the value of “too many” spectral bands and the concomitant wealth of spatial information that was amenable to information extraction by statistical techniques. One of them was Eyal Ben-Dor who pioneered soil analyses using hyperspectral imaging and who is one of the contributors to this volume.

ix

x

Foreword to the First Edition

As the quality of AVIRIS data grew, manifested in increasing SNR, an ever-increasing amount of information could be extracted from the data. This quality was reflected in the increasing number of nearly noiseless principal components that could be obtained from the data or, in other words, its dimensionality. The explosive advances in desktop computing made possible the application of image processing and statistical analyses that revolutionized the uses of hyperspectral imaging. Joe Boardman and others at the University of Colorado developed what has become the ENVI software package to make possible the routine analysis of hyperspectral image data using “unmixing techniques” to derive the relative abundance of surface materials on a pixel-by-pixel basis. Many of the analysis techniques discussed in this volume, such as band selection and various indices, are rooted in principal components analysis. The eigenvector loadings or factors indicate which spectral bands are the most heavily weighted allowing others to be discarded to reduce the noise contribution. As sensors become better, more information will be extractable and fewer bands will be discarded. This is the beauty of hyperspectral imaging, allowing the choice of the number of eigenvectors to be used for a particular problem. Computing power has reached such a high level that it is no longer necessary to choose a subset of bands just to minimize the computational time. As regression techniques such as PLS (partial least squares) become increasingly adopted to relate a particular vegetation parameter to reflectance spectra, it must be remembered that the quality of the calibration model is a function of both the spectra and the reference measurement. With spectral measurements of organic and inorganic compounds under laboratory conditions, we have found that a poor model with a low coefficient of determination (r2) is most often associated with inaccurate reference measurements, leading to the previously intuitive conclusion that “spectra don’t lie.” Up to this point, AVIRIS has provided the bulk of high-quality hyperspectral image data but on an infrequent basis. Although Hyperion has provided some time series data, there is no hyperspectral imager yet in orbit that is capable of providing routine, high-quality images of the whole Earth on a consistent basis. The hope is that in the next decade, HyspIRI will be providing VNIR and SWIR hyperspectral images every 3 weeks and multispectral thermal data every week. This resource will revolutionize the field of vegetation remote sensing since so much of the useful information is bound up in the seasonal growth cycle. The combination of the spectral, spatial, and temporal dimensions will be ripe for the application of statistical techniques and the results will be extraordinary. Dr. Alexander F. H. Goetz PhD Former Chairman and Chief Scientist ASD Inc. 2555 55th St. #100 Boulder, CO 80301, USA 303-444-6522 ext. 108 Fax 303-444-6825 www.asdi.com

REFERENCES Goetz, A. F. H., 2009, Three decades of hyperspectral imaging of the Earth: A personal view, Remote Sensing of Environment, 113, S5–S16. Goetz, A.F.H., G. Vane, J. Solomon and B.N. Rock, 1985, Imaging spectrometry for Earth remote sensing, Science, 228, 1147–1153.

BIOGRAPHICAL SKETCH Dr. Goetz is one of the pioneers in hyperspectral remote sensing and certainly needs no introduction. Dr. Goetz started his career working on spectroscopic reflectance and emittance studies of the Moon and Mars. He was a principal investigator of Apollo-8 and Apollo-12 multispectral photography

Foreword to the First Edition

xi

studies. Later, he turned his attention to remote sensing of Planet Earth working in collaboration with Dr. Gene Shoemaker to map geology of Coconino County (Arizona) using Landsat-1 data and went on to be an investigator in further Landsat, Skylab, Shuttle, and EO-1 missions. At NASA/JPL he pioneered field spectral measurements and initiated the development of hyperspectral imaging. He spent 21 years on the faculty of the University of Colorado, Boulder, and retired in 2006 as an Emeritus Professor of Geological Sciences and an Emeritus Director of Center for the Study of Earth from Space. Since then, he has been Chairman and Chief Scientist of ASD Inc. a company that has provided more than 850 research laboratories in over 60 countries with field spectrometers. Dr. Goetz is now retired. His foreword was written for the first edition and I have retained it in consultation with him to get a good perspective on the development of hyperspectral remote sensing.

Foreword to the Second Edition The publication of the four-volume set, Hyperspectral Remote Sensing of Vegetation, second edition, is a landmark effort in providing an important, valuable, and timely contribution that summarizes the state of spectroscopy-based understanding of the Earth’s terrestrial and near shore environments. Imaging spectroscopy has had 35 years of development in data processing and analysis methods. Today’s researchers are eager to use data produced by hyperspectral imagers and address important scientific issues from agricultural management to global environmental stewardship. The field started with development of the Jet Propulsion Lab’s Airborne Imaging Spectrometer in 1983 that measured across the reflected solar infrared spectrum with 128 spectral bands. This technology was quickly followed in 1987 by the more capable Advanced Visible Infrared Imaging Spectrometer (AVIRIS), which has flown continuously since this time (albeit with multiple upgrades). It has 224 spectral bands covering the 400–2500 nm range with 10 nm wavelength bands and represents the “gold standard” of this technology. In the years since then, progress toward a hyperspectral satellite has been disappointingly slow. Nonetheless, important and significant progress in understanding how to analyze and understand spectral data has been achieved, with researchers focused on developing the concepts, analytical methods, and spectroscopic understanding, as described throughout these four volumes. Much of the work up to the present has been based on theoretical analysis or from experimental studies at the leaf level from spectrometer measurements and at the canopy level from airborne hyperspectral imagers. Although a few hyperspectral satellites have operated over various periods in the 2000s, none have provided systematic continuous coverage required for global mapping and time series analysis. An EnMap document compiled the past and near-term future hyperspectral satellites and those on International Space Station missions (EnMap and GRSS Technical Committee 2017). Of the hyperspectral imagers that have been flown, the European Space Agency’s CHRIS (Compact High Resolution Imaging Spectrometer) instrument on the PROBA-1 (Project for On-Board Autonomy) satellite and the Hyperion sensor on the NASA technology demonstrator, Earth Observing-1 platform (terminated in 2017). Each has operated for 17 years and have received the most attention from the science community. Both collect a limited number of images per day, and have low data quality relative to today’s capability, but both have open data availability. Other hyperspectral satellites with more limited access and duration include missions from China, Russia, India, and the United States. We are at a threshold in the availability of hyperspectral imagery. There are many hyperspectral missions planned for launch in the next 5 years from China, Italy, Germany, India, Japan, Israel, and the United States, some with open data access. The analysis of the data volumes from this proliferation of hyperspectral imagers requires a comprehensive reference resource for professionals and students to turn to in order to understand and correctly and efficiently use these data. This four-volume set is unique in compiling in-depth understanding of calibration, visualization, and analysis of data from hyperspectral sensors. The interest in this technology is now widespread, thus, applications of hyperspectral imaging cross many disciplines, which are truly international, as is evident by the list of authors of the chapters in these volumes, and the number of countries planning to operate a hyperspectral satellite. At least some of the hyperspectral satellites announced and expected to be launched in this decade (such as the HyspIRI-like satellite approved for development by NASA with a launch in the 2023 period) will provide high-fidelity narrow-wavelength bands, covering the full reflected solar spectrum, at moderate (30 m pixels) to high spatial resolution. These instruments will have greater radiometric range, better SNR, pointing accuracy, and reflectance calibration than past instruments, and will collect data from many countries and parts of the world that have not previously been available. Together, these satellites will produce an unprecedented flow of information about the physiological functioning (net primary production, evapotranspiration, and even direct measurements related to respiration), biochemical characteristics (from spectral indices xiii

xiv

Foreword to the Second Edition

and from radiative transfer first principle methods), and direct measurements of the distributions of plant and soil biodiversity of the terrestrial and coastal environments of the Earth. This four-volume set presents an unprecedented range and scope of information on hyperspectral data analysis and applications written by leading authors in the field. Topics range from sensor characteristics from ground-based platforms to satellites, methods of data analysis to characterize plant functional properties related to exchange of gases CO2, H2O, O2, and biochemistry for pigments, N cycle, and other molecules. How these data are used in applications range from precision agriculture to global change research. Because the hundreds of bands in the full spectrum includes information to drive detection of these properties, the data is useful at scales from field applications to global studies. Volume I has three sections and starts with an introduction to hyperspectral sensor systems. Section II focuses on sensor characteristics from ground-based platforms to satellites, and how these data are used in global change research, particularly in relation to agricultural crop monitoring and health of natural vegetation. Section III provides five chapters that deal with the concept of spectral libraries to identify crops and spectral traits, and for phenotyping for plant breeding. It addresses the development of spectral libraries, especially for agricultural crops and one for soils. Volume II expands on the first volume, focusing on use of hyperspectral indices and image classification. The volume begins with an explanation of how narrow-band hyperspectral indices are determined, often from individual spectral absorption bands but also from correlation matrices and from derivative spectra. These are followed by chapters on statistical approaches to image classification and a chapter on methods for dealing with “big data.” The last half of this volume provides five chapters focused on use of vegetation indices for quantifying and characterizing photosynthetic pigments, leaf nitrogen concentrations or contents, and foliar water content measurements. These chapters are particularly focused on applications for agriculture, although a chapter addresses more heterogeneous forest conditions and how these patterns relate to monitoring health and production. The first half of Volume III focuses on biophysical and biochemical characterization of vegetation properties that are derived from hyperspectral data. Topics include ecophysiological functioning and biomass estimates of crops and grasses, indicators of photosynthetic efficiency, and stress detection. The chapter addresses biophysical characteristics across different spatial scales while another chapter examines spectral and spatial methods for retrieving biochemical and biophysical properties of crops. The chapters in the second half of this volume are focused on identification and discrimination of species from hyperspectral data and use of these methods for rapid phenotyping of plant breeding trials. Lastly, two chapters evaluate tree species identification, and another provides examples of mapping invasive species. Volume IV focuses on six areas of advanced applications in agricultural crops. The first considers detection of plant stressors including nitrogen deficiency and excess heavy metals and crop disease detection in precision farming. The second addresses global patterns of crop water productivity and quantifying litter and invasive species in arid ecosystems. Phenological patterns are examined while others focus on multitemporal data for mapping patterns of phenology. The third area is focused on applications of land cover mapping in different forest, wetland, and urban applications. The fourth topic addresses hyperspectral measurements of wildfires, and the fifth evaluates use of continuity vegetation index data in global change applications. And lastly, the sixth area examines use of hyperspectral data to understand the geologic surfaces of other planets. Susan L. Ustin Professor and Vice Chair, Dept. Land, Air and Water Resources Associate Director, John Muir Institute University of California Davis California, USA

Foreword to the Second Edition

xv

REFERENCE EnMap Ground Segment Team and GSIS GRSS Technical committee, December, 2017. Spaceborne Imaging Spectroscopy Mission Compilation. DLR Space Administration and the German Federal Ministry of Economic Affairs and Technology. http://www.enmap.org/sites/default/files/pdf/Hyperspectral_EO_ Missions_2017_12_21_FINAL4.pdf

BIOGRAPHICAL SKETCH Dr. Susan L. Ustin is currently a Distinguished Professor of Environmental and Resource Sciences in the Department of Land, Air, and Water Resources, University of California Davis, Associate Director of the John Muir Institute, and is Head of the Center for Spatial Technologies and Remote Sensing (CSTARS) at the same university. She was trained as a plant physiological ecologist but began working with hyperspectral imagery as a post-doc in 1983 with JPL’s AIS program. She became one of the early adopters of hyperspectral remote sensing which has now extended over her entire academic career. She was a pioneer in the development of vegetation analysis using imaging spectrometery, and is an expert on ecological applications of this data. She has served on numerous NASA, NSF, DOE, and the National Research Council committees related to spectroscopy and remote sensing. Among recognitions for her work, she is a Fellow of the American Geophysical Union and received an honorary doctorate from the University of Zurich. She has published more than 200 scientific papers related to ecological remote sensing and has worked with most of the Earth-observing U.S. airborne and spaceborne systems.

Preface This seminal book on Hyperspectral Remote Sensing of Vegetation (Second Edition, 4 Volume Set), published by Taylor and Francis Inc.\CRC Press is an outcome of over 2 years of effort by the editors and authors. In 2011, the first edition of Hyperspectral Remote Sensing of Vegetation was published. The book became a standard reference on hyperspectral remote sensing of vegetation amongst the remote sensing community across the world. This need and resulting popularity demanded a second edition with more recent as well as more comprehensive coverage of the subject. Many advances have taken place since the first edition. Further, the first edition was limited in scope in the sense it covered some very important topics and missed equally important topics (e.g., hyperspectral library of agricultural crops, hyperspectral pre-processing steps and algorithms, and many others). As a result, a second edition that brings us up-to-date advances in hyperspectral remote sensing of vegetation was required. Equally important was the need to make the book more comprehensive, covering an array of subjects not covered in the first edition. So, my coeditors and myself did a careful research on what should go into the second edition. Quickly, the scope of the second edition expanded resulting in an increasing number of chapters. All of this led to developing the seminal book: Hyperspectral Remote Sensing of Vegetation, Second Edition, 4 Volume Set. The four volumes are: Volume I: Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation Volume II: Hyperspectral Indices and Image Classifications for Agriculture and Vegetation Volume III: Biophysical and Biochemical Characterization and Plant Species Studies Volume IV: Advanced Applications in Remote Sensing of Agricultural Crops and Natural Vegetation The goal of the book was to bring in one place collective knowledge of the last 50 years of advances in hyperspectral remote sensing of vegetation with a target audience of wide spectrum of scientific community, students, and professional application practitioners. The book documents knowledge advances made in applying hyperspectral remote sensing technology in the study of terrestrial vegetation that include agricultural crops, forests, rangelands, and wetlands. This is a very practical offering about a complex subject that is rapidly advancing its knowledge-base. In a very practical way, the book demonstrates the experience, utility, methods, and models used in studying terrestrial vegetation using hyperspectral data. The four volumes, with a total of 48 chapters, are divided into distinct themes. • Volume I: There are 14 chapters focusing on hyperspectral instruments, spectral libraries, and methods and approaches of data handling. The chapters extensively address various preprocessing steps and data mining issues such as the Hughes phenomenon and overcoming the “curse of high dimensionality” of hyperspectral data. Developing spectral libraries of crops, vegetation, and soils with data gathered from hyperspectral data from various platforms (ground-based, airborne, spaceborne), study of spectral traits of crops, and proximal sensing at field for phenotyping are extensively discussed. Strengths and limitations of hyperspectral data of agricultural crops and vegetation acquired from different platforms are discussed. It is evident from these chapters that the hyperspectral data provides opportunities for great advances in study of agricultural crops and vegetation. However, it is also clear from these chapters that hyperspectral data should not be treated as panacea to every limitation of multispectral broadband data such as from Landsat or Sentinel series of satellites. The hundreds or thousands of hyperspectral narrowbands (HNBs) as well as carefully selected hyperspectral vegetation indices (HVIs) will help us make significant xvii

xviii

Preface

advances in characterizing, modeling, mapping, and monitoring vegetation biophysical, biochemical, and structural quantities. However, it is also important to properly understand hyperspectral data and eliminate redundant bands that exist for every application and to optimize computing as well as human resources to enable seamless and efficient handling enormous volumes of hyperspectral data. Special emphasis is also put on preprocessing and processing of Earth Observing-1 (EO-1) Hyperion, the first publicly available hyperspectral data from space. These methods, approaches, and algorithms, and protocols set the stage for upcoming satellite hyperspectral sensors such as NASA’s HyspIRI and Germany’s EnMAP. • Volume II: There are 10 chapters focusing on hyperspectral vegetation indices (HVIs) and image classification methods and techniques. The HVIs are of several types such as: (i) twoband derived, (ii) multi-band-derived, and (iii) derivative indices derived. The strength of the HVIs lies in the fact that specific indices can be derived for specific biophysical, biochemical, and plant structural quantities. For example, you have carotenoid HVI, anthocyanin HVI, moisture or water HVI, lignin HVI, cellulose HVI, biomass or LAI or other biophysical HVIs, red-edge based HVIs, and so on. Further, since these are narrowband indices, they are better targeted and centered at specific sensitive wavelength portions of the spectrum. The strengths and limitations of HVIs in a wide array of applications such as leaf nitrogen content (LNC), vegetation water content, nitrogen content in vegetation, leaf and plant pigments, anthocyanin’s, carotenoids, and chlorophyll are thoroughly studied. Image classification using hyperspectral data provides great strengths in deriving more classes (e.g., crop species within a crop as opposed to just crop types) and increasing classification accuracies. In earlier years and decades, hyperspectral data classification and analysis was a challenge due to computing and data handling issues. However, with the availability of machine learning algorithms on cloud computing (e.g., Google Earth Engine) platforms, these challenges have been overcome in the last 2–3 years. Pixel-based supervised machine learning algorithms like the random forest, and support vector machines as well as objectbased algorithms like the recursive hierarchical segmentation, and numerous others methods (e.g., unsupervised approaches) are extensively discussed. The ability to process petabyte volume data of the planet takes us to a new level of sophistication and makes use of data such as from hyperspectral sensors feasible over large areas. The cloud computing architecture involved with handling massively large petabyte-scale data volumes are presented and discussed. • Volume III: There are 11 chapters focusing on biophysical and biochemical characterization and plant species studies. A number of chapters in this volume are focused on separating and discriminating agricultural crops and vegetation of various types or species using hyperspectral data. Plant species discrimination and classification to separate them are the focus of study using vegetation such as forests, invasive species in different ecosystems, and agricultural crops. Performance of hyperspectral narrowbands (HNBs) and hyperspectral vegetation indices (HVIs) when compared with multispectral broadbands (MBBs) and multispectral broadband vegetation indices (BVIs) are presented and discussed. The vegetation and agricultural crops are studied at various scales, and their vegetation functional properties diagnosed. The value of digital surface models in study of plant traits as complementary\supplementary to hyperspectral data has been highlighted. Hyperspectral bio-indicators to study photosynthetic efficiency and vegetation stress are presented and discussed. Studies are conducted using hyperspectral data across wavelengths (e.g., visible, near-infrared, shortwave-infrared, mid-infrared, and thermal-infrared). • Volume IV: There are 15 chapters focusing on specific advanced applications of hyperspectral data in study of agricultural crops and natural vegetation. Specific agricultural crop applications include crop management practices, crop stress, crop disease, nitrogen application, and presence of heavy metals in soils and related stress factors. These studies discuss biophysical and biochemical quantities modeled and mapped for precision farming,

Preface

xix

hyperspectral narrowbands (HNBs), and hyperspectral vegetation indices (HVIs) involved in assessing nitrogen in plants, and the study of the impact of heavy metals on crop health and stress. Vegetation functional studies using hyperspectral data presented and discussed include crop water use (actual evapotranspiration), net primary productivity (NPP), gross primary productivity (GPP), phenological applications, and light use efficiency (LUE). Specific applications discussed under vegetation functional studies using hyperspectral data include agricultural crop classifications, machine learning, forest management studies, pasture studies, and wetland studies. Applications in fire assessment, modeling, and mapping using hyperspectral data in the optical and thermal portions of the spectrum are presented and discussed. Hyperspectral data in global change studies as well as in outer planet studies have also been discussed. Much of the outer planet remote sensing is conducted using imaging spectrometer and hence the data preprocessing and processing methods of Earth and that of outer planets have much in common and needs further examination. The chapters are written by leading experts in the global arena with each chapter: (a) focusing on specific applications, (b) reviewing existing “state-of-art” knowledge, (c) highlighting the advances made, and (d) providing guidance for appropriate use of hyperspectral data in study of vegetation and its numerous applications such as crop yield modeling, crop biophysical and biochemical property characterization, and crop moisture assessment. The four-volume book is specifically targeted on hyperspectral remote sensing as applied to terrestrial vegetation applications. This is a big market area that includes agricultural croplands, study of crop moisture, forests, and numerous applications such as droughts, crop stress, crop productivity, and water productivity. To the knowledge of the editors, there is no comparable book, source, and/or organization that can bring this body of knowledge together in one place, making this a “must buy” for professionals. This is clearly a unique contribution whose time is now. The book highlights include: 1. Best global expertise on hyperspectral remote sensing of vegetation, agricultural crops, crop water use, plant species detection, crop productivity and water productivity mapping, and modeling; 2. Clear articulation of methods to conduct the work. Very practical; 3. Comprehensive review of the existing technology and clear guidance on how best to use hyperspectral data for various applications; 4. Case studies from a variety of continents with their own subtle requirements; and 5. Complete solutions from methods to applications inventory and modeling. Hyperspectral narrowband spectral data, as discussed in various chapters of this book, are fast emerging as practical most advanced solutions in modeling and mapping vegetation. Recent research has demonstrated the advances and great value made by hyperspectral data, as discussed in various chapters in: (a) quantifying agricultural crops as to their biophysical and harvest yield characteristics, (b) modeling forest canopy biochemical properties, (c) establishing plant and soil moisture conditions, (d) detecting crop stress and disease, (e) mapping leaf chlorophyll content as it influences crop production, (f) identifying plants affected by contaminants such as arsenic, and (g) demonstrating sensitivity to plant nitrogen content, and (h) invasive species mapping. The ability to significantly better quantify, model, and map plant chemical, physical, and water properties is well established and has great utility. Even though these accomplishments and capabilities have been reported in various places, the need for a collective “knowledge bank” that links these various advances in one place is missing. Further, most scientific papers address specific aspects of research, failing to provide a comprehensive assessment of advances that have been made nor how the professional can bring those advances to their work. For example, deep scientific journals report practical applications of hyperspectral

xx

Preface

narrowbands yet one has to canvass the literature broadly to obtain the pertinent facts. Since several papers report this, there is a need to synthesize these findings so that the reader gets the correct picture of the best wavebands for their practical applications. Also, studies do differ in exact methods most suited for detecting parameters such as crop moisture variability, chlorophyll content, and stress levels. The professional needs this sort of synthesis and detail to adopt best practices for their own work. In years and decades past, use of hyperspectral data had its challenges especially in handling large data volumes. That limitation is now overcome through cloud-computing, machine learning, deep learning, artificial intelligence, and advances in knowledge in processing and applying hyperspectral data. This book can be used by anyone interested in hyperspectral remote sensing that includes advanced research and applications, such as graduate students, undergraduates, professors, practicing professionals, policy makers, governments, and research organizations.

Dr. Prasad S. Thenkabail, PhD Editor-in-Chief Hyperspectral Remote Sensing of Vegetation, Second Edition, Four Volume Set

Acknowledgments This four-volume Hyperspectral Remote Sensing of Vegetation book (second edition) was made possible by sterling contributions from leading professionals from around the world in the area of hyperspectral remote sensing of vegetation and agricultural crops. As you will see from list of authors and coauthors, we have an assembly of “who is who” in hyperspectral remote sensing of vegetation who have contributed to this book. They wrote insightful chapters, that are an outcome of years of careful research and dedication, to make the book appealing to a broad section of readers dealing with remote sensing. My gratitude goes to (mentioned in no particular order; names of lead authors of the chapters are shown in bold): Drs. Fred Ortenberg (Technion–Israel Institute of Technology, Israel), Jiaguo Qi (Michigan State University, USA), Angela Lausch (Helmholtz Centre for Environmental Research, Leipzig, Germany), Andries B. Potgieter (University of Queensland, Australia), Muhammad Al-Amin Hoque (University of Queensland, Australia), Andreas Hueni (University of Zurich, Switzerland), Eyal Ben-Dor (Tel Aviv University, Israel), Itiya Aneece (United States Geological Survey, USA), Sreekala Bajwa (University of Arkansas, USA), Antonio Plaza (University of Extremadura, Spain), Jessica J. Mitchell (Appalachian State University, USA), Dar Roberts (University of California at Santa Barbara, USA), Quan Wang (Shizuoka University, Japan), Edoardo Pasolli (University of Trento, Italy), (Nanjing University of Science and Technology, China), Anatoly Gitelson (University of Nebraska- Lincoln, USA), Tao Cheng (Nanjing Agricultural University, China), Roberto Colombo (University of Milan-Bicocca, Italy), Daniela Stroppiana (Institute for Electromagnetic Sensing of the Environment, Italy), Yongqin Zhang (Delta State University, USA), Yoshio Inoue (National Institute for Agro-Environmental Sciences, Japan), Yafit Cohen (Institute of Agricultural Engineering, Israel), Helge Aasen (Institute of Agricultural Sciences, ETH Zurich), Elizabeth M. Middleton (NASA, USA), Yongqin Zhang (University of Toronto, Canada), Yan Zhu (Nanjing Agricultural University, China), Lênio Soares Galvão (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Matthew L. Clark (Sonoma State University, USA), Matheus Pinheiro Ferreira (University of Paraná, Curitiba, Brazil), Ruiliang Pu (University of South Florida, USA), Scott C. Chapman (CSIRO, Australia), Haibo Yao (Mississippi State University, USA), Jianlong Li (Nanjing University, China), Terry Slonecker (USGS, USA), Tobias Landmann (International Centre of Insect Physiology and Ecology, Kenya), Michael Marshall (University of Twente, Netherlands), Pamela Nagler (USGS, USA), Alfredo Huete (University of Technology Sydney, Australia), Prem Chandra Pandey (Banaras Hindu University, India), Valerie Thomas (Virginia Tech., USA), Izaya Numata (South Dakota State University, USA), Elijah W. Ramsey III (USGS, USA), Sander Veraverbeke (Vrije Universiteit Amsterdam and University of California, Irvine), Tomoaki Miura (University of Hawaii, USA), R. G. Vaughan (U.S. Geological Survey, USA), Victor Alchanatis (Agricultural research Organization, Volcani Center, Israel), Dr. Narumon Wiangwang (Royal Thai Government, Thailand), Pedro J. Leitão (Humboldt University of Berlin, Department of Geography, Berlin, Germany), James Watson (University of Queensland, Australia), Barbara George-Jaeggli (ETH Zuerich, Switzerland), Gregory McLean (University of Queensland, Australia), Mark Eldridge (University of Queensland, Australia), Scott C. Chapman (University of Queensland, Australia), Kenneth Laws (University of Queensland, Australia), Jack Christopher (University of Queensland, Australia), Karine Chenu (University of Queensland, Australia), Andrew Borrell (University of Queensland, Australia), Graeme L. Hammer (University of Queensland, Australia), David R. Jordan (University of Queensland, Australia), Stuart Phinn (University of Queensland, Australia), Lola Suarez (University of Melbourne, Australia), Laurie A. Chisholm (University of Wollongong, Australia), Alex Held (CSIRO, Australia), S. Chabrillant (GFZ German Research Center for Geosciences, Germany), José A. M. Demattê (University of São Paulo, Brazil), Yu Zhang (North Dakota State University, USA), Ali Shirzadifar (North Dakota State University, USA), Nancy F. Glenn (Boise State University, USA), Kyla M. Dahlin (Michigan State xxi

xxii

Acknowledgments

University, USA), Nayani Ilangakoon (Boise State University, USA), Hamid Dashti (Boise State University, USA), Megan C.  Maloney (Appalachian State University, USA), Subodh Kulkarni (University of Arkansas, USA), Javier Plaza (University of Extremadura, Spain), Gabriel Martin (University of Extremadura, Spain), Segio Sánchez (University of Extremadura, Spain), Wei Wang (Nanjing Agricultural University, China), Xia Yao (Nanjing Agricultural University, China), Busetto Lorenzo (Università Milano-Bicocca), Meroni Michele (Università Milano-Bicocca), Rossini Micol (Università Milano-Bicocca), Panigada Cinzia (Università Milano-Bicocca), F. Fava (Università degli Studi di Sassari, Italy), M. Boschetti (Institute for Electromagnetic Sensing of the Environment, Italy), P. A. Brivio (Institute for Electromagnetic Sensing of the Environment, Italy), K. Fred Huemmrich (University of Maryland, Baltimore County, USA), Yen-Ben Cheng (Earth Resources Technology, Inc., USA), Hank A. Margolis (Centre d’Études de la Forêt, Canada), Yafit Cohen (Agricultural research Organization, Volcani Center, Israel), Kelly Roth (University of California at Santa Barbara, USA), Ryan Perroy (University of Wisconsin-La Crosse, USA), Ms. Wei Wang (Nanjing Agricultural University, China), Dr. Xia Yao (Nanjing Agricultural University, China), Keely L. Roth (University of California, Santa Barbara, USA), Erin B. Wetherley (University of California at Santa Barbara, USA), Susan K. Meerdink (University of California at Santa Barbara, USA), Ryan L. Perroy (University of Wisconsin-La Crosse, USA), B. B. Marithi Sridhar (Bowling Green University, USA), Aaryan Dyami Olsson (Northern Arizona University, USA), Willem Van Leeuwen (University of Arizona, USA), Edward Glenn (University of Arizona, USA), José Carlos Neves Epiphanio (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Fábio Marcelo Breunig (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Antônio Roberto Formaggio (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Amina Rangoonwala (IAP World Services, Lafayette, LA), Cheryl Li (Nanjing University, China), Deghua Zhao (Nanjing University, China), Chengcheng Gang (Nanjing University, China), Lie Tang (Mississippi State University, USA), Lei Tian (Mississippi State University, USA), Robert Brown (Mississippi State University, USA), Deepak Bhatnagar (Mississippi State University, USA), Thomas Cleveland (Mississippi State University, USA), Hiroki Yoshioka (Aichi Prefectural University, Japan), T. N. Titus (U.S. Geological Survey, USA), J. R. Johnson (U.S. Geological Survey, USA), J. J. Hagerty (U.S. Geological Survey, USA), L. Gaddis (U.S. Geological Survey, USA), L. A. Soderblom (U.S. Geological Survey, USA), and P. Geissler (U.S. Geological Survey, USA), Jua Jin (Shizuoka University, Japan), Rei Sonobe (Shizuoka University, Japan), Jin Ming Chen (Shizuoka University, Japan), Saurabh Prasad (University of Houston, USA), Melba M. Crawford (Purdue University, USA), James C. Tilton (NASA Goddard Space Flight Center, USA), Jin Sun (Nanjing University of Science and Technology, China), Yi Zhang (Nanjing University of Science and Technology, China), Alexei Solovchenko (Moscow State University, Moscow), Yan Zhu, (Nanjing Agricultural University, China), Dong Li (Nanjing Agricultural University, China), Kai Zhou (Nanjing Agricultural University, China), Roshanak Darvishzadeh (University of Twente, Enschede, The Netherlands), Andrew Skidmore (University of Twente, Enschede, The Netherlands), Victor Alchanatis (Institute of Agricultural Engineering, The Netherlands), Georg Bareth (University of Cologne, Germany), Qingyuan Zhang (Universities Space Research Association, USA), Petya K. E. Campbell (University of Maryland Baltimore County, USA), and David R. Landis (Global Science & Technology, Inc., USA), José Carlos Neves Epiphanio (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Fábio Marcelo Breunig (Universidade Federal de Santa Maria [UFSM], Brazil), and Antônio Roberto Formaggio (Instituto Nacional de Pesquisas Espaciais [INPE], Brazil), Cibele Hummel do Amaral (Federal University of Viçosa, in Brazil), Gaia Vaglio Laurin (Tuscia University, Italy), Raymond Kokaly (U.S. Geological Survey, USA), Carlos Roberto de Souza Filho (University of Ouro Preto, Brazil), Yosio Edemir Shimabukuro (Federal Rural University of Rio de Janeiro, Brazil), Bangyou Zheng (CSIRO, Australia), Wei Guo (The University of Tokyo, Japan), Frederic Baret (INRA, France), Shouyang Liu (INRA, France), Simon Madec (INRA, France), Benoit Solan (ARVALIS, France), Barbara George-Jaeggli (University of Queensland, Australia), Graeme L. Hammer (University of Queensland, Australia), David R. Jordan (University of Queensland, Australia), Yanbo Huang (USDA, USA), Lie Tang (Iowa State

Acknowledgments

xxiii

University, USA), Lei Tian (University of Illinois. USA), Deepak Bhatnagar (USDA, USA), Thomas E. Cleveland (USDA, USA), Dehua ZHAO (Nanjing University, USA), Hannes Feilhauer (University of Erlangen-Nuremberg, Germany), Miaogen Shen (Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China), Jin Chen (College of Remote Sensing Science and Engineering, Faculty of Geographical Science, Beijing Normal University, Beijing, China), Suresh Raina (International Centre of Insect Physiology and Ecology, Kenya and Pollination services, India), Danny Foley (Northern Arizona University, USA), Cai Xueliang (UNESCO-IHE, Netherlands), Trent Biggs (San Diego State University, USA), Werapong Koedsin (Prince of Songkla University, Thailand), Jin Wu (University of Hong Kong, China), Kiril Manevski (Aarhus University, Denmark), Prashant K. Srivastava (Banaras Hindu University, India), George P. Petropoulos (Technical University of Crete, Greece), Philip Dennison (University of Utah, USA), Ioannis Gitas (University of Thessaloniki, Greece), Glynn Hulley (NASA Jet Propulsion Laboratory, California Institute of Technology, USA), Olga Kalashnikova, (NASA Jet Propulsion Laboratory, California Institute of Technology, USA), Thomas Katagis (University of Thessaloniki, Greece), Le Kuai (University of California, USA), Ran Meng (Brookhaven National Laboratory, USA), Natasha Stavros (California Institute of Technology, USA). Hiroki Yoshioka (Aichi Prefectural University, Japan), My two coeditors, Professor John G. Lyon and Professor Alfredo Huete, have made outstanding contribution to this four-volume Hyperspectral Remote Sensing of Vegetation book (second edition). Their knowledge of hyperspectral remote sensing is enormous. Vastness and depth of their understanding of remote sensing in general and hyperspectral remote sensing in particular made my job that much easier. I have learnt a lot from them and continue to do so. Both of them edited some or all of the 48 chapters of the book and also helped structure chapters for a flawless reading. They also significantly contributed to the synthesis chapter of each volume. I am indebted to their insights, guidance, support, motivation, and encouragement throughout the book project. My coeditors and myself are grateful to Dr. Alexander F. H. Goetz and Prof. Susan L. Ustin for writing the foreword for the book. Please refer to their biographical sketch under the respective foreword written by these two leaders of Hyperspectral Remote Sensing. Both the forewords are a must read to anyone studying this four-volume Hyperspectral Remote Sensing of Vegetation book (second edition). They are written by two giants who have made immense contribution to the subject and I highly recommend that the readers read them. I am blessed to have had the support and encouragement (professional and personal) of my U.S. Geological Survey and other colleagues. In particular, I would like to mention Mr. Edwin Pfeifer (late), Dr. Susan Benjamin, Dr. Dennis Dye, and Mr. Larry Gaffney. Special thanks to Dr. Terrence Slonecker, Dr. Michael Marshall, Dr. Isabella Mariotto, and Dr. Itiya Aneece who have worked closely with me on hyperspectral research over the years. Special thanks are also due to Dr. Pardhasaradhi Teluguntla, Mr. Adam Oliphant, and Dr. Muralikrishna Gumma who have contributed to my various research efforts and have helped me during this book project directly or indirectly. I am grateful to Prof. Ronald B. Smith, professor at Yale University who was instrumental in supporting my early hyperspectral research at the Yale Center for Earth Observation (YCEO), Yale University. Opportunities and guidance I received in my early years of remote sensing from Prof. Andrew D. Ward, professor at the Ohio State University, Prof. John G. Lyon, former professor at the Ohio State University, and Mr. Thiruvengadachari, former Scientist at the National Remote Sensing Center (NRSC), Indian Space Research Organization, India, is gratefully acknowledged. My wife (Sharmila Prasad) and daughter (Spandana Thenkabail) are two great pillars of my life. I am always indebted to their patience, support, and love. Finally, kindly bear with me for sharing a personal story. When I started editing the first edition in the year 2010, I was diagnosed with colon cancer. I was not even sure what the future was and how long I would be here. I edited much of the first edition soon after the colon cancer surgery and during and after the 6 months of chemotherapy—one way of keeping my mind off the negative thoughts. When you are hit by such news, there is nothing one can do, but to be positive, trust your

xxiv

Acknowledgments

doctors, be thankful to support and love of the family, and have firm belief in the higher spiritual being (whatever your beliefs are). I am so very grateful to some extraordinary people who helped me through this difficult life event: Dr. Parvasthu Ramanujam (surgeon), Dr. Paramjeet K. Bangar (Oncologist), Dr. Harnath Sigh (my primary doctor), Dr. Ram Krishna (Orthopedic Surgeon and family friend), three great nurses (Ms. Irene, Becky, Maryam) at Banner Boswell Hospital (Sun City, Arizona, USA), courage-love-patience-prayers from my wife, daughter, and several family members, friends, and colleagues, and support from numerous others that I have not named here. During this phase, I learnt a lot about cancer and it gave me an enlightened perspective of life. My prayers were answered by the higher power. I learnt a great deal about life—good and bad. I pray for all those with cancer and other patients that diseases one day will become history or, in the least, always curable without suffering and pain. Now, after 8 years, I am fully free of colon cancer and was able to edit the four-volume Hyperspectral Remote Sensing of Vegetation book (second edition) without the pain and suffering that I went through when editing the first edition. What a blessing. These blessings help us give back in our own little ways. To realize that it is indeed profound to see the beautiful sunrise every day, the day go by with every little event (each with a story of their own), see the beauty of the sunset, look up to the infinite universe and imagine on its many wonders, and just to breathe fresh air every day and enjoy the breeze. These are all many wonders of life that we need to enjoy, cherish, and contemplate.

Dr. Prasad S. Thenkabail, PhD Editor-in-Chief Hyperspectral Remote Sensing of Vegetation

Editors Prasad S. Thenkabail, Research Geographer-15, U.S. Geological Survey (USGS), is a world-recognized expert in remote sensing science with multiple major contributions in the field sustained over more than 30 years. He obtained his PhD from the Ohio State University in 1992 and has over 140+ peer-reviewed scientific publications, mostly in major international journals. Dr. Thenkabail has conducted pioneering research in the area of hyperspectral remote sensing of vegetation and in that of global croplands and their water use in the context of food security. In hyperspectral remote sensing he has done cutting-edge research with wide implications in advancing remote sensing science in application to agriculture and vegetation. This body of work led to more than ten peer-reviewed research publications with high impact. For example, a single paper [1] has received 1000+ citations as at the time of writing (October 4, 2018). Numerous other papers, book chapters, and books (as we will learn below) are also related to this work, with two other papers [2,3] having 350+ to 425+ citations each. In studies of global croplands in the context of food and water security, he has led the release of the world’s first Landsat 30-m derived global cropland extent product. This work demonstrates a “paradigm shift” in how remote sensing science is conducted. The product can be viewed in full resolution at the web location www.croplands.org. The data is already widely used worldwide and is downloadable from the NASA\USGS LP DAAC site [4]. There are numerous major publication in this area (e.g. [5,6]). Dr. Thenkabail’s contributions to series of leading edited books on remote sensing science places him as a world leader in remote sensing science advances. He edited three-volume Remote Sensing Handbook published by Taylor and Francis, with 82 chapters and more than 2000 pages, widely considered a “magnus opus” standard reference for students, scholars, practitioners, and major experts in remote sensing science. Links to these volumes along with endorsements from leading global remote sensing scientists can be found at the location give in note [7]. He has recently completed editing Hyperspectral Remote Sensing of Vegetation published by Taylor and Francis in four volumes with 50 chapters. This is the second edition is a follow-up on the earlier single-volume Hyperspectral Remote Sensing of Vegetation [8]. He has also edited a book on Remote Sensing of Global Croplands for Food Security (Taylor and Francis) [9]. These books are widely used and widely referenced in institutions worldwide. Dr. Thenkabail’s service to remote sensing community is second to none. He is currently an editor-in-chief of the Remote Sensing open access journal published by MDPI; an associate editor of the journal Photogrammetric Engineering and Remote Sensing (PERS) of the American Society of Photogrammetry and Remote Sensing (ASPRS); and an editorial advisory board member of the International Society of Photogrammetry and Remote Sensing (ISPRS) Journal of Photogrammetry and Remote Sensing. Earlier, he served on the editorial board of Remote Sensing of Environment for many years (2007–2017). As an editor-in-chief of the open access Remote Sensing MDPI journal from 2013 to date he has been instrumental in providing leadership for an online publication that did not even have a impact factor when he took over but is now one of the five leading remote sensing international journals, with an impact factor of 3.244. Dr. Thenkabail has led remote sensing programs in three international organizations: International Water Management Institute (IWMI), 2003–2008; International Center for Integrated Mountain Development (ICIMOD), 1995–1997; and International Institute of Tropical Agriculture (IITA), xxv

xxvi

Editors

1992–1995. He has worked in more than 25+ countries on several continents, including East Asia (China), S-E Asia (Cambodia, Indonesia, Myanmar, Thailand, Vietnam), Middle East (Israel, Syria), North America (United States, Canada), South America (Brazil), Central Asia (Uzbekistan), South Asia (Bangladesh, India, Nepal, and Sri Lanka), West Africa (Republic of Benin, Burkina Faso, Cameroon, Central African Republic, Cote d’Ivoire, Gambia, Ghana, Mali, Nigeria, Senegal, and Togo), and Southern Africa (Mozambique, South Africa). During this period he has made major contributions and written seminal papers on remote sensing of agriculture, water resources, inland valley wetlands, global irrigated and rain-fed croplands, characterization of African rainforests and savannas, and drought monitoring systems. The quality of Dr. Thenkabail’s research is evidenced in the many awards, which include, in 2015, the American Society of Photogrammetry and Remote Sensing (ASPRS) ERDAS award for best scientific paper in remote sensing (Marshall and Thenkabail); in 2008, the ASPRS President’s Award for practical papers, second place (Thenkabail and coauthors); and in 1994, the ASPRS Autometric Award for outstanding paper (Thenkabail and coauthors). His team was recognized by the Environmental System Research Institute (ESRI) for “special achievement in GIS” (SAG award) for their Indian Ocean tsunami work. The USGS and NASA selected him to be on the Landsat Science Team for a period of five years (2007–2011). Dr. Thenkabail is regularly invited as keynote speaker or invited speaker at major international conferences and at other important national and international forums every year. He has been principal investigator and/or has had lead roles of many pathfinding projects, including the ∼5 million over five years (2014–2018) for the global food security support analysis data in the 30-m (GFSAD) project (https://geography.wr.usgs.gov/science/croplands/) funded by NASA MEaSUREs (Making Earth System Data Records for Use in Research Environments), and projects such as Sustain and Manage America’s Resources for Tomorrow (waterSMART) and characterization of Eco-Regions in Africa (CERA).

REFERENCES 1. Thenkabail, P.S., Smith, R.B., and De-Pauw, E. 2000b. Hyperspectral vegetation indices for determining agricultural crop characteristics. Remote Sensing of Environment, 71:158–182. 2. Thenkabail, P.S., Enclona, E.A., Ashton, M.S., Legg, C., and Jean De Dieu, M. 2004. Hyperion, IKONOS,  ALI, and ETM+ sensors in the study of African rainforests. Remote Sensing of Environment, 90:23–43. 3. Thenkabail, P.S., Enclona, E.A., Ashton, M.S., and Van Der Meer, V. 2004. Accuracy assessments of hyperspectral waveband performance for vegetation analysis applications. Remote Sensing of Environment, 91(2–3):354–376. 4. https://lpdaac.usgs.gov/about/news_archive/release_gfsad_30_meter_cropland_extent_products 5. Thenkabail, P.S. 2012. Guest Editor for Global Croplands Special Issue. Photogrammetric Engineering and Remote Sensing, 78(8). 6. Thenkabail, P.S., Knox, J.W., Ozdogan, M., Gumma, M.K., Congalton, R.G., Wu, Z., Milesi, C., Finkral, A., Marshall, M., Mariotto, I., You, S. Giri, C. and Nagler, P. 2012. Assessing future risks to agricultural productivity, water resources and food security: how can remote sensing help? Photogrammetric Engineering and Remote Sensing, August 2012 Special Issue on Global Croplands: Highlight Article. 78(8):773–782. IP-035587. 7. https://www.crcpress.com / Remote-Sensing-Handbook---Three-Volume-Set/Thenkabail/p/ book/9781482218015 8. https://www.crcpress.com/Hyperspectral-Remote-Sensing-of-Vegetation/Thenkabail-Lyon/p/ book/9781439845370 9. ht t p s: //w w w.c r c p r e ss.c om / Remo t e - Sen si ng- of- Globa l- Cr opla nd s -fo r- Fo o d - Se cu r it y/ Thenkabail-Lyon-Turral-Biradar/p/book/9781138116559

Editors

xxvii

John G. Lyon, educated at Reed College in Portland, OR and the University of Michigan in Ann Arbor, has conducted scientific and engineering research and carried out administrative functions throughout his career. He was formerly the Senior Physical Scientist (ST) in the US Environmental Protection Agency’s Office of Research and Development (ORD) and Office of the Science Advisor in Washington, DC, where he co-led work on the Group on Earth Observations and the USGEO subcommittee of the Committee on Environment and Natural Resources and research on geospatial issues in the agency. For approximately eight years, he was director of ORD’s Environmental Sciences Division, which conducted research on remote sensing and geographical information system (GIS) technologies as applied to environmental issues including landscape characterization and ecology, as well as analytical chemistry of hazardous wastes, sediments, and ground water. He previously served as professor of civil engineering and natural resources at Ohio State University (1981–1999). Professor Lyon’s own research has led to authorship or editorship of a number of books on wetlands, watershed, and environmental applications of GIS, and accuracy assessment of remote sensor technologies. Alfredo Huete leads the Ecosystem Dynamics Health and Resilience research program within the Climate Change Cluster (C3) at the University of Technology Sydney, Australia. His main research interest is in using remote sensing to study and analyze vegetation processes, health, and functioning, and he uses satellite data to observe land surface responses and interactions with climate, land use activities, and extreme events. He has more than 200 peerreviewed journal articles, including publication in such prestigious journals as Science and Nature. He has over 25 years’ experience working on NASA and JAXA mission teams, including the NASAEOS MODIS Science Team, the EO-1 Hyperion Team, the JAXA GCOM-SGLI Science Team, and the NPOESS-VIIRS advisory group. Some of his past research involved the development of the soil-adjusted vegetation index (SAVI) and the enhanced vegetation index (EVI), which became operational satellite products on MODIS and VIIRS sensors. He has also studied tropical forest phenology and Amazon forest greening in the dry season, and his work was featured in a National Geographic television special entitled “The Big Picture.” Currently, he is involved with the Australian Terrestrial Ecosystem Research Network (TERN), helping to produce national operational phenology products; as well as the AusPollen network, which couples satellite sensing to better understand and predict pollen phenology from allergenic grasses and trees.

Contributors Itiya P. Aneece Western Geographic Science Center United States Geological Survey Flagstaff, Arizona Sreekala G. Bajwa Department of Agricultural and Biosystems Engineering North Dakota State University Fargo, North Dakota E. Ben-Dor Department of Geography Porter School of Environment and Earth Science Tel-Aviv University Tel-Aviv, Israel Andrew Borrell Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia S. Chabrillat Remote Sensing Helmholtz Centre Potsdam GFZ German Research Centre for Geosciences Potsdam, Germany Scott C. Chapman CSIRO Agriculture Flagship and School of Agriculture and Food Sciences The University of Queensland Queensland, Australia

Jack Christopher Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia Kyla M. Dahlin Department of Geography, Environment, and Spatial Sciences Michigan State University East Lansing, Michigan Hamid Dashti Department of Geosciences Boise State University Boise, Idaho José A. M. Demattê Department of Soil Science College of Agriculture “Luiz de Queiroz” University of São Paulo São Paulo, Brazil Mark Eldridge Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia Barbara George-Jaeggli Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia

Karine Chenu Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia

and

Laurie A. Chisholm School of Earth an Environmental Sciences University of Wollongong Wollongong, Australia

Nancy F. Glenn Department of Geosciences Boise State University Boise, Idaho

Department of Agriculture and Fisheries Agri-Science Queensland Queensland, Australia

xxix

xxx

Graeme L. Hammer Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia Alex Held CSIRO-Land and Water Commonwealth Scientific and Industrial Research Organisation Canberra, Australia Muhammad Al-Amin Hoque Remote Sensing Research Centre School of Earth and Environmental Sciences University of Queensland Brisbane, Australia and Department of Geography and Environment Jagannath University Dhaka, Bangladesh Andreas Hueni Geography Department Remote Sensing Laboratories University of Zurich Zurich, Switzerland Alfredo Huete School of Life Sciences University of Technology Sydney New South Wales, Australia Nayani Ilangakoon Department of Geosciences Boise State University Boise, Idaho Yoshio Inoue Agro-Ecosystem Informatics Research National Institute for Agro-Environmental Sciences Ibaraki, Japan David R. Jordan Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia

Contributors

Angela Lausch Department Computational Landscape Ecology Helmholtz Centre for Environmental Research—UFZ Leipzig, Germany and Department of Geography Humboldt University of Berlin Berlin, Germany Kenneth Laws Department of Agriculture and Fisheries Agri-Science Queensland Queensland, Australia Pedro J. Leitão Department of Geography Humboldt University of Berlin Berlin, Germany and Department of Landscape Ecology and Environmental System Analysis Technische Universität Braunschweig Braunschweig, Germany John G. Lyon American Society for Photogrammetry and Remote Sensing Chantilly, Virginia Megan C. Maloney Department of Geography Appalachian State University Boone, North Carolina Gabriel Martín Instituto Superior Técnico Technical University of Lisbon Lisbon, Portugal Gregory McLean Department of Agriculture and Fisheries Agri-Science Queensland Queensland, Australia Jessica J. Mitchell Montana Natural Heritage Program Spatial Analysis Lab University of Montana Missoula, Montana

xxxi

Contributors

Fred Ortenberg Technion Israel Institute of Technology Haifa, Israel Stuart Phinn Remote Sensing Research Centre School of Earth and Environmental Sciences University of Queensland Brisbane, Australia Antonio Plaza Department of Technology of Computers and Communications Hyperspectral Computing Laboratory University of Extremadura Cáceres, Spain Javier Plaza Department of Technology of Computers and Communications University of Extremadura Badajoz (UNEX), Spain Andries B. Potgieter Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia Jiaguo Qi Department of Geography Center for Global Change and Earth Observations Michigan State University East Lansing, Michigan Sergio Sánchez Department of Technology of Computers and Communications Hyperspectral Computing Laboratory University of Extremadura Cáceres, Spain

Alimohammad Shirzadifar Department of Agricultural and Biosystems Engineering North Dakota State University Fargo, North Dakota Terrance Slonecker Eastern Geographic Science Center United States Geological Survey Reston, Virginia Lola Suarez Department of Infrastructure Engineering University of Melbourne Melbourne, Australia Prasad S. Thenkabail Western Geographic Science Center United States Geological Survey Flagstaff, Arizona James Watson Queensland Alliance for Agriculture and Food Innovation (QAAFI) University of Queensland Brisbane, Australia Narumon Wiangwang Royal Thai Government Department of Fisheries Information Technology Center Bangkok, Thailand Yu Zhang Department of Agricultural and Biosystems Engineering North Dakota State University Fargo, North Dakota

Acronyms and Abbreviations ρ σ δ ν R2 A/CHEAT AC ACORN ADEOS AISA ALI ALS ANN AOP API ASCII ASD ASI ASO ASTER ASU ATCOR ATLAS ATREM AVHRR AVIRIS BRDF BSC C COV CAO CBERS-2 CDA CEOS CHRIS CHV CLM CMF CNES CNN CO2 CoETP COVER% CP CPU

Correlation Variance Bending Vibration Stretching Vibration Coefficient of Determination Auscover/Curtin Hyperion Enhancement and Atmospheric Correction Technique Atmospheric Corrector Atmospheric Correction Now Advanced Earth Observing Satellite Airborne Image Spectrometer for Applications Advanced Land Imager Airborne Laser Scanning Artificial Neural Network Airborne Observation Platform Application Programming Interface American Standard Code for Information Interchange Analytical Spectral Devices Inc. Agenzia Spaziale Italiana Airborne Snow Observatory Advanced Spaceborne Thermal Emission and Reflection Radiometer Arizona State University Atmospheric and Topographic Correction Advanced Topographic Laser Altimeter System Atmospheric Removal Advanced Very High-Resolution Radiometer Airborne Visible and Infrared Imaging Spectrometer Bidirectional Reflectance Distribution Function Biological species concept Carbon Covariance Carnegie Airborne Observatory China-Brazil Earth Resources Satellite Canonical Discriminant Analysis Committee on Earth Observation Satellite Compact High Resolution Imaging Spectrometer Convex Hull Volume Community Land Model Color Matching Functions Centre National d’Etudes Spatiales Convoluted Neural Network Carbon Dioxide ARC Centre of Excellence for Translational Photosynthesis Crop cover percentage Crude protein (%) Central Processing Unit xxxiii

xxxiv

CSR CTIC CWA CWC CWSI D DAAC DAIS DART DBH DESIS DLR DISORT DN DNA DOAS DoD DSM DTM DWEL E EBVs ECOSTRESS ECOsystem ED EFFORT EGU ELD EMP EnMAP ENVI-IDL ENVI ENVISAT EO-1 EO EOS EOSDIS ERDAS ERDAS EROS ESA ESDIS ESMO ESMs ETM+ EUFAR EVI FAO FEDM FLAASH FLDA FLEX FLiES

Acronyms and Abbreviations

Competitive-, stress, and ruderal strategy types of plants Cross-Track Illumination Correction Continuous Wavelet Analysis Canopy Water Content Crop Water Stress Index Derivative Distributed Active Archive Center Digital Airborne Imaging Spectrometer Discrete Anisotropic Radiative Transfer Diameter at Breast Height Earth Sensing Imaging Spectrometer Discrete Ordinate Radiative Transfer Digital Numbers Deoxyribonucleic acid Differential Optical Absorption Spectroscopy Department of Defense Digital Surface Model Digital Terrain Model Dual-Wavelength Echidna LiDAR Radiance Essential Biodiversity Variables Spaceborne Thermal Radiometer Experiment on Space Station Ecosystem Demography Empirical Flag Field Optimal Reflectance Transformation European Geoscience Union Effective Leaf Density Extended Morphological Profile Environmental Mapping and Analysis Programme Environment for Visualizing Images-Interactive Data Language Environment for Visualizing Images Environmental Satellite Earth Observing-1 Earth Observations Earth Observing System Earth Observing System Data and Information System Earth Resource Data Analysis System Earth Resource Development Assessment System Earth Resources Observation and Science European Space Agency Earth Science Data and Information System Earth Science Mission Operations Earth System Models Enhanced Thematic Mapper Plus European Facility for Airborne Research Echidna Validation Instrument Food and agriculture organization of the United Nations Frequent Domain Electro Magnetic Fast-Line-of-Sight Atmospheric Analysis of Hypercubes Fisher’s Linear Discriminant Analysis Fluorescence Explorer Forest Light Environment Simulator

Acronyms and Abbreviations

FloX FNAI FNIR FORMOSAT FOV FTHSI FTIR FWF FWHM G-LiHT GA GB GECKO GEDI GEE GEO GEOEYE-1 and 2 GERIS GIS GLAS GLI GOME GORT GPR GPS GxExM H HDF HICO HISUI HRS HSI HSR HRS HTTP HTV HVI HyMAP Hyperion HyPLANT HyspIRI ICA ICAMM ICARE ICESat IKONOS

xxxv

Fluorescence Box—dual field of view spectroradiometer Florida Natural Areas Inventory Far near-infrared (1100–1300 nm) Satellite operated by Taiwanese National Space Organization NSPO Data marketed by SPOT Image Field of View Fourier Transform Hyperspectral Imager Fourier Transform Infrared Spectroscopy Full Waveform Full Width at Half Maximum (pulse width) Goddard’s LiDAR, Hyperspectral and Thermal Airborne Imager Genetic Algorithm Gigabyte Genetic by environment characterization (via) kinetic observation Global Ecosystem Dynamics Investigation Google Earth Engine BON Group on Earth Observation Biodiversity Observation Network Providing data in 0.25–1.65 m resolution Geophysical and Environmental Research Imaging Spectrometer Geographic Information System Geoscience Laser Altimeter System Global Imager Global Ozone Monitoring Experiment Geometric Optical-Radiative Transfer Ground Penetrating Radar Global Positioning System Genotype by environment by management Entropy Hierarchical Data Format Hyperspectral Imager for the Coastal Ocean Hyperspectral Imager SUIte Hyperspectral remote sensing Hyperspectral Imagers High Spatial Resolution Hyperspectral Remote Sensing High-throughput phenotyping H-2 Transfer Vehicle Hyperspectral vegetation indices Hyperspectral MAPping sensor First spaceborne hyperspectral sensor onboard Earth Observing-1(EO-1) Airborne imaging spectrometer with a fluorescence module Hyperspectral InfraRed Imager Independent Component Analysis Independent Component Analysis-based Mixed Model International Conference on Airborne Research for the Environment Ice, Cloud, and Land Elevation Satellite Very High Spatial Resolution (sub-meter to 5-m) satellite operated by GeoEye

xxxvi

INSAR IR IRS-1C/D-LISS IRS-P6/AWiFS IS ISS JAXA JHU JPL JRBP KFD KOMFOSAT KPLSR L LA LAI LAN Landsat-1, 2, 3 MSS Landsat-4, 5 TM Landsat-7 ETM+ LANDSAT-TM LANDSAT LDA LiDAR LMA LNA LOD LOWTRAN LOW LOWTRAN LSMs LUE LUI LULC LVIS LWC LWIR MEI MESMA MF MI MIA MLC MLC MLP MLR MLRA MNDVI MNF MODIS MODTRAN Mrad

Acronyms and Abbreviations

Interferometric Synthetic-Aperture Radar Infrared Indian Remote Sensing Satellite/Linear Imaging Self Scanner Indian Remote Sensing Satellite/Advanced Wide Field Sensor Imaging Spectroscopy International Space Station Japan Aerospace Exploration Agency Johns Hopkins University Jet Propulsion Laboratory Jasper Ridge Biological Preserve Kernel Fisher Discriminant Korean multipurpose satellite. Data marketed by SPOT Image Kernel-based Partial Least Square Regression Irradiance Leaf Area Leaf area index (m2/m2) Leaf Angle Multispectral Scanner Thematic Mapper Enhanced Thematic Mapper Plus Land Satellite Thematic Mapper Sensor Land Satellite Multispectral Sensor Linear Discriminant Analysis Light Detection and Ranging Leaf Mass per Area Leaf nitrogen accumulation Linked open data Resolution model for predicting atmosphere Transitions Low Resolution Atmospheric Transmittance Land Surface Models Light use efficiency Land-use intensity Land Use, Land Cover Land, Vegetation, and Ice Sensor Leaf Water Content Longwave Infrared Management, equipment, and infrastructure Multiple End-member Spectral Mixture Analysis Matched Filter Mutual Information Mutual Information Analysis Maximum Likelihood Classification Maximum Likelihood Classifier Multi-Layer Perceptron Multiple Linear Regression Machine Learning Regression Algorithms Modified Normalized Differential Vegetation Index Minimum Noise Fraction Moderate Resolution Imaging Spectroradiometer Moderate Resolution Atmospheric Transmittance Milliradians

xxxvii

Acronyms and Abbreviations

MS MSC MWIR N N NANO NASA JPL NASA NASA NDI nDSM NDVI NDWI NeCTAR NEON NIR NIRS NN NOAA NPP NRL NSA NSC NSMI OLS OM OMI OO OSP P PAI PC PCA PFTs PLNTHT PLS PLSR POS PP PRF PRI PRISMA PROBA PSC QSM QSMT QUICKBIRD RAPID EYE REP RESOURCESAT RF RGB

Multispectral Morphological species concept Mid-Wave Infrared Nitrogen (%) Number (of pixels, bands, and so on) Headwall Nano-Hyperspec camera Jet Propulsion Laboratory National Aeronautics and Space Administration Normalized Difference Index Normalized Digital Surface Model Normalized Difference Vegetation Index Normalized Difference Water Index National eResearch collaboration tools and resources project National Ecological Observatory Network Near-Infrared Near-Infrared Spectroscopy Neural Network National Oceanic and Atmospheric Administration Net Primary Productivity Naval Research Laboratory Normalized Spectral Area Non-Structural Carbohydrates Normalized Soil Moisture Index Ordinary Least Squares Organic matter Ozone Monitoring Instrument Ocean Optics USB2000+ point spectrometer Orthogonal Subspace Projection Phosphorous Plant Area Index Principal Component Principal Components Analysis Plant Functional Types Plant height (mm) Partial Least Squares Partial Least Square Regression Penetrating Optical Sensor Projection Pursuit Pulse Repetition Rate Photochemical reflectance index PRecursore IperSpettrale della Missione Applicativa Project for On-Board Autonomy Phylogenetic species concept Quantitative Structure Models Quantitative Spectral Matching Technique Satellite from DigitalGlobe, a private company in the USA Satellite constellation from Rapideye, a German company Red-edge Position Determination Satellite launched by India Random Forest Red-Green-Blue

xxxviii

RMSE RPAS RPD RS-ST/STV-C RS RSRC RTM S SALCA SAM SAVI SBS SBUV SCIAMACHY SCP SDAR SFF SfM SFS SID SIF SIS SMA SMARTS SMGM SNR SOC SPEAR (ENVI tool) SPECIM SPECPR SPOT ST STV SVAT SVH SVM SVR SWIR TAU TCARI TES TIR TLS TOMS UAS UAV UFL UMV UNFCCC UQ

Acronyms and Abbreviations

Root Mean Square Error Remotely Piloted Airborne System Ratio of Prediction to Deviation Remote sensing-spectral trait/spectral trait variation-concept Remote sensing Remote Sensing Research Centre Radiative Transfer Model Scatter Matrix Salford Advanced Laser Canopy Analyser Spectral Angle Mapper Soil-Adjusted Vegetating Index Sequential Backward Selection Solar Backscatter Ultraviolet Scanning Imaging Absorption Spectrometer for Atmospheric Cartography Sorghum conversion program Spectral Diversity Area Relationships Spectral Feature Fitting Structure from Motion Sequential Forward Selection Spectral Information Divergence Solar induced chlorophyll fluorescence Spectral Information System Spectral Mixture Analysis Simple Model of the Radiative Transfer of Sunshine Soil Moisture Gaussian Model Signal-to-Noise Ratio Soil Organic Carbon Spectral Processing Exploitation and Analysis Resource SPECtral IMaging SPECtrum Processing Routines Système Pour l’Observation de la Terre Spectral traits Spectral trait variations Soil-Vegetation-Atmosphere-Transfer Spectral Variation Hypothesis Support Vector Machine Support Vector Regression Shortwave Infrared Tel Aviv University Transformed Chlorophyll Absorption in Reflectance Index Tropospheric Emission Spectrometer Thermal Infrared Terrestrial Laser Scanning Total Ozone Mapping Spectrometer Unmanned Aircraft System Unmanned Aerial Vehicle Ultraviolet Fluorescence LiDAR Unmanned Vehicle United Nations Framework Convention on Climate Change University of Queensland

xxxix

Acronyms and Abbreviations

USAD USGS UV Vc VCA Vcmax VDMs VI VIS -NIR VIS VMEMSMA VNIR VSL VSWIR W WASMA WBM WFIS WORLDVIEW WR WSC WSN WT YI

United States Department of Agriculture United States Geological Survey Ultraviolet Rate that rubisco works in chloroplasts Vertex Component Analysis Maximum rate for a given leaf (temperature dependent) Vegetation Dynamical Models Vegetation Indices Visible and Near-Infrared Visible Variable Multiple EndMember Spectral Mixture Analysis Visible and Near-Infrared Vegetation Spectral Library Visible and shortwave infrared Water Wavelength Adaptive Spectral Mixture Analysis Wet biomass (kg\m2) Wide Field-of-view Imaging Spectrometer DigitalGlobe’s Earth imaging satellite White reference World Soil Congress Wireless sensor networks Wavelet Transform Yellowness Index

Section I Introduction to Hyperspectral Remote Sensing of Agricultural Crops and Vegetation

1

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops Prasad S. Thenkabail, John G. Lyon, and Alfredo Huete

CONTENTS 1.1 1.2 1.3 1.4

Introduction and Rationale ....................................................................................................... 3 Hyperspectral Remote Sensing of Vegetation and Agricultural Crops .................................. 12 Hyperspectral Data Composition for Study of Vegetation and Agricultural Crops ............... 16 Methods and Approaches of Hyperspectral Data Analysis for Vegetation and Agricultural Crops .................................................................................................................. 16 1.4.1 Lambda (λ1) versus Lambda (λ2) Plots....................................................................... 18 1.4.2 Principal Component Analysis ................................................................................... 19 1.4.3 Other Hyperspectral Data Mining Algorithms .......................................................... 19 1.4.4 Optimal Hyperspectral Narrow-Bands: Hyperspectral Vegetation Indices (HVIs) to Study Vegetation and Crop Biophysical and Biochemical Properties .......20 1.4.4.1 Hyperspectral Two-Band Vegetation Index (HTBVI) .................................20 1.4.4.2 Hyperspectral Multiple-Band Models (HMBMs) ........................................ 21 1.4.4.3 Hyperspectral Derivative Greenness Vegetation Indices (HDGVIs) .......... 22 1.4.4.4 Hyperspectral Hybrid Vegetation Indices .................................................... 23 1.4.5 Other Methods of Hyperspectral Data Analysis ........................................................24 1.4.6 Broad-Band Vegetation Index Models ........................................................................25 1.5 Separating Vegetation Classes and Agricultural Crops Using Hyperspectral Narrow-Band Data ..................................................................................................................25 1.5.1 Class Separability Using Unique Hyperspectral Narrow-Bands ................................25 1.5.2 Class Separability Using Statistical Methods .............................................................26 1.5.3 Accuracy Assessments of Vegetation and Crop Classification Using Hyperspectral Narrow-Bands ..................................................................................... 27 1.6 Optimal Hyperspectral Narrow-Bands in the Study of Vegetation and Agricultural Crops ...28 1.7 Conclusions ............................................................................................................................. 31 Acknowledgment ............................................................................................................................. 31 References ........................................................................................................................................ 32

1.1

INTRODUCTION AND RATIONALE

In the recent past, hyperspectral remote sensing has gained great utility for a variety of applications. At last, one can do remotely what chemists have done in laboratories with spectra. It is now possible to be diagnostic in sensing species and plant communities and do so in a direct and informed manner. The true promise of remote sensing for plant studies is realized with application of modern tools and analysis. Hyperspectral data can provide significant improvements in spectral information content when compared with traditional broad-band analysis. We now can address a variety of phenomena and processes only dreamed of. A number of thoughtful investigators have demonstrated robust 3

4

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

applications, and a list of these includes: (a) detecting plant stress (Thenkabail et al., 2004b), (b) measuring the chlorophyll content of plants (Blackburn and Ferwerda, 2008), (c) identifying small differences in percent of green vegetation cover (Chen et  al., 2008), (d) extracting biochemical variables such as nitrogen and lignin (Blackburn and Ferwerda, 2008, Chan and Paelinckx, 2008, Dalponte et al., 2009; Houborg and Boegh, 2008), (e) discriminating land cover types (Thenkabail et  al., 2004a), (f) detecting crop moisture variations (Colombo et  al., 2008), (g) sensing subtle variations in leaf pigment concentrations (Yang et  al., 2009, Zhao et  al., 2007, Blackburn and Ferwerda, 2008), (h) modeling biophysical and yield characteristics of agricultural crops (Thenkabail et al., 2000, 2002, Houborg et al., 2009), (i) improving the detection of changes in sparse vegetation (Lyon et al., 1998), and (j) assessing absolute water content in plant leaves (Jollineau and Howarth, 2008). This is a fairly detailed list but not exhaustive. It gives the reader a measure of the current, proven experimental capabilities and operational applications. It also elevates the mind in thinking of new, ambitious applications that are attainable by driven investigators. The spectral properties of vegetation are strongly determined by their biophysical and biochemical attributes such as leaf area index (LAI), the amount of live biomass and senesced biomass, moisture content, pigments (e.g., chlorophyll), and spatial arrangement of structures (Asner, 1998, Hill, 2004). We are now capable of measuring those phenomena and processes, and harness them in testing hypotheses and valuable applications on a variety of ecosystems. For example, the assessment of biophysical and biochemical properties of vegetation such as rangelands (Darvishzadeh et al., 2008b, c), agricultural crops (Thenkabail et al., 2000, Thenkabail et al., 2002, 2004a,b, Darvishzadeh et al., 2008a), and weeds (Thenkabail et al., 2004a) are essential for evaluating productivity, providing information needed for local farmers and institutions, and assessing grazing potential for livestock. Even though remote sensing has been recognized as a reliable method for estimating various biophysical and biochemical vegetation variables, the existing broad-band sensors were inadequate or supplied limited information for the purpose (Thenkabail, 2003, Thenkabail et  al., 2004a,b, Vaiphasa et  al., 2007, Maire et  al., 2008, Yao et  al., 2010). Clearly, broad-bands have known limitations in providing adequate information on properties such as crop growth stage, crop type differentiation, generation of agricultural crop statistics, identification of forest type and species identification, characterizing complex-forest versus non-forest interactions, and detailed mapping of land-cover classes and all to serve diverse scientific and other user communities (Thenkabail et al., 2002, 2004a). These limitations of broad-band analysis are illustrated by vegetation indices (VIs) which saturate beyond a certain level of biomass and leaf area index (LAI) (Thenkabail et al., 2000). For example, VIs typically increase over an LAI range from 0 to about 3–5 before an asymptote is reached. While extremely useful over the years, the upper limit of this sensitivity apparently differs among vegetation types and can only be driven so far to a solution for a given application. Saturation is more pronounced for planophile canopies (Atzberger, 2004). However, compared with erectophile canopies of the same LAI, planophile canopies are less influenced by soil brightness variations (Darvishzadeh et al., 2008a). In contrast, hyperspectral datasets allow identification of features, allow direct measurement of canopy variables such as biochemical content (e.g., chlorophyll, nitrogen, lignin), forest species, chemistry distribution, timber volumes, water conditions (Gong et al. 1995, Carter, 1998, Blackburn and Ferwerda, 2008), as well as biophysical (e.g., leaf area index, biomass) and yield characteristics (Thenkabail et al., 2002, Thenkabail et al., 2004a, Zhao et al., 2007, Galvão et al., 2009). Hyperspectral sensors gather near-continuous spectra from imaging spectrometers such as the National Aeronautics and Space Administration (NASA) designed Airborne Visible-InfraRed Imaging Spectrometer (AVIRIS) and Compact Airborne Spectrographic Imager (CASI). This new generation of spaceborne sensors offer tremendous improvements in spatial, spectral, radiometric, and temporal resolutions as well as improvements in optics and mechanics when compared with older generations of sensors (Table 1.1). The promise and potential of hyperspectral narrow-band sensors for a wide array of Earth resource applications motivated the design and launch of spaceborne

Spatial (meters)

1000

250, 500, 1000

56 × 79

30

Sensor

1. AVHRR

2. MODIS

3. Landsat-1, 2, 3 MSS

4. Landsat-4, 5 TM

8

6

4

7

12

11

36/7

4

Spectral (#)

Band Range (µm)

Band Widths (µm)

0.10 0.375 0.38 0.65 0.7

0.036 0.02 0.02 0.02 0.02 0.05

0.84–0.876 0.459–0.479 0.545–0.565 1.23–1.25 1.63–1.65 2.11–2.16 0.1 0.1 0.1 0.3 0.07 0.80 0.60 0.14

0.5–0.6 0.6–0.7 0.7–0.8 0.8–1.1 0.45–0.52 0.52–0.60 0.63–0.69 0.76–0.90

C. Multispectral Broad-Band Sensors

0.05

0.62–0.67

B. Coarse Resolution Narrow-Band Sensors

0.58–0.68 0.725–1.1 3.55–3.93 10.30–10.95 10.95–11.65

A. Coarse Resolution Broad-Band Sensors

Radiometric (bit)

1843 1555 1047 1970 1843 1555 1047

1970

2053 1719.8 447.4 227.4 86.7

974.3

1528.2

1390 1410 1510 0 0

Irradiance (W m−2 sr−1 µm−1)

11.1

2.26

0.16, 0.04, 0.01 0.16, 0.04, 0.01

0.01

Data Points (# per hectares)

TABLE 1.1 Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

(Continued)

16

16

daily

daily

Frequency of Revisit (days)

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops 5

8

30

15, 30, 90

5. Landsat-7 ETM+

6. ASTER

15

Spectral (#)

Spatial (meters)

Sensor

12

8

8

Radiometric (bit)

0.80 0.60 0.150 0.200 2.5 0.2 0.38 0.11 0.06 0.1 0.1 0.1 0.04 0.04 0.05 0.07 0.07 0.35 0.35 0.35 0.7 0.7

0.52–0.60 0.63–0.69 0.50–0.75 0.75–0.90 10.0–12.5 1.75–1.55 0.52–0.90 (p) 0.52–0.63 0.63–0.69 0.76–0.86 0.76–0.86 1.60–1.70 2.145–2.185 2.185–2.225 2.235–2.285 2.295–2.365 2.360–2.430 8.125–8.475 8.475–8.825 8.925–9.275 10.25–10.95 10.95–11.65

Band Widths (µm) 0.19 2.10 0.25 0.65

Band Range (µm) 1.55–1.74 10.4–12.5 2.08–2.35 0.45–0.52

1546.0 1117.6 1117.6 232.5 80.32 74.96 69.20 59.82 57.32 0 0 0 0 0

1843 1555 1047 227.1 0 1368 1352.71 1846.9

227.1 0 80.53 1970

Irradiance (W m−2 sr−1 µm−1)

44.4, 11.1, 1.23

44.4, 11.1

Data Points (# per hectares)

TABLE 1.1 (continued) Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

(Continued)

16

16

Frequency of Revisit (days)

6 Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Spatial (meters)

30

2.5–20

23.5

23.5

56

Sensor

7. ALI

8. SPOT-1 -2 -3 -4

9. IRS-1C

10. IRS-1

11. IRS-P6-AWiFS

4

15

15

15

10

Spectral (#)

10

8

8

16

12

Radiometric (bit) 0.048–0.69 (p) 0.433–0.453 0.450–0.515 0.425–0.605 0.633–0.690 0.775–0.805 0.845–0.890 1.200–1.300 1.550–1.750 2.080–2.350 0.50–0.59 0.61–0.68 0.79–0.89 1.5–1.75 0.51–0.73 (p) 0.52–0.59 0.62–0.68 0.77–0.86 1.55–1.70 0.5–0.75 (P) 0.52–0.59 0.62–0.68 0.77–0.86 1.55–1.70 0.5–0.75 (P) 0.52–0.59 0.62–0.68

Band Range (µm) 0.64 0.20 0.65 0.80 0.57 0.30 0.45 1.00 2.00 2.70 0.09 0.07 0.1 0.25 0.22 0.07 0.06 0.09 0.15 0.25 0.07 0.06 0.09 0.15 0.25 0.07 0.06

Band Widths (µm) 1747.8600 1849.5 1985.0714 1732.1765 1485.2308 1134.2857 948.36364 439.61905 223.39024 78.072727 1858 1575 1047 234 1773 1851.1 1583.8 1102.5 240.4 1627.1 1852.1 1577.38 1096.7 240.4 1603.9 1857.7 1556.4

Irradiance (W m−2 sr−1 µm−1)

3.19

18.1

18.1

1600, 25

11.1

Data Points (# per hectares)

TABLE 1.1 (continued) Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

(Continued)

16

16

16

3–5

16

Frequency of Revisit (days)

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops 7

1134 cm2 @ 1.2 m Nadir view 18° Field of view 60

60

14. ASD spectroradiometer

16. HyspIRI TIR

15. HyspIRI VSWIR

30

20 m pan, 20 m MS 5 m pan, 20 m MS

Spatial (meters)

13. Hyperion

–4

–3B –3

12. CBERS –2

Sensor

8

2100 bands 1 nm width between 400–2500 nm 210

220 (196a)

Spectral (#)

16

16

16

16

11

0.12

0.77–0.89

Band Widths (µm) 0.09 0.15 0.22 0.07 0.07 0.06

Band Range (µm) 0.77–0.86 1.55–1.70 0.51–0.73 0.45–0.52 0.52–0.59 0.63–0.69

7 bands in 7500–12,000 nm 3000–5000 nm (3980 nm center)

210 bands in 380–2500 nm

196 effective Calibrated bands VNIR (band 8–57) 427.55–925.85 nm SWIR (band 79–224) 932.72–2395.53 nm 2100 effective bands

10 nm wide (approx.) for all 210 bands 7 bands in 7500– 12,000 nm and 1 band in

1 nm wide (approx.) in 400–2500 nm

10 nm wide (approx.) for all 196 bands

D. Hyperspectral Narrow-Band Sensors

Radiometric (bit)

See data in Neckel and Labs (1984). Plot it and obtain values for Hyperion bands See data in Neckel and Labs (1984). Plot it See data in Neckel and Labs (1984). Plot it

See data in Neckel and Labs (1984). Plot it and obtain values for Hyperion bands

1664.3

1082.4 239.84 1934.03 1787.10 1587.97 1069.21

Irradiance (W m−2 sr−1 µm−1)

2.77

2.77

88,183

11.1

400, 25

25, 25

Data Points (# per hectares)

TABLE 1.1 (continued) Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

(Continued)

5

19

5–16

16

Frequency of Revisit (days)

8 Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Spatial (meters)

1–4

0.61–2.44

5.8

6.5

0.55 2–8

Sensor

17. IKONOS

18. QUICKBIRD

19. RESOURSESAT

20. RAPID EYE -A -E

21. WORLDVIEW 22. FORMOSAT-2

1 5

5

3

4

4

Spectral (#)

11 11

12

10

11

11

Radiometric (bit) Band Widths (µm) 0.71 0.89 0.66 0.96 0.07 0.08 0.06 0.13 0.07 0.06 0.09 0.07 0.07 0.05 0.04 0.12 0.06 0.07 0.08 0.06 0.14 0.45

0.445–0.516 0.506–0.595 0.632–0.698 0.757–0.853 0.45–0.52 0.52–0.60 0.63–0.69 0.76–0.89 0.52–0.59 0.62–0.68 0.77–0.86 0.44–0.51 0.52–0.59 0.63–0.68 0.69–0.73 0.77–0.89 0.45–0.51 0.45–0.52 0.52–0.60 0.63–0.69 0.76–0.90 0.45–0.90 (p)

E. Hyperspatial Broad-Band Sensors

Band Range (µm)

1743.12 1485.23 1041.28 1450

1930.9 1854.8 1156.5 1156.9 1381.79 1924.59 1843.08 1574.77 1853.6 1581.6 1114.3 1979.33 1752.33 1499.18 1343.67 1039.88 1996.77 1974.93

Irradiance (W m−2 sr−1 µm−1)

40,000 2500, 156.25

236.7

33.64

14,872, 625

10,000, 625

Data Points (# per hectares)

TABLE 1.1 (continued) Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

(Continued)

1.7–5.9 daily

1–2

24

5

5

Frequency of Revisit (days)

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops 9

5

Spectral (#) 10

Band Range (µm) 0.5–0.9 0.45–0.52 0.52–0.6 0.63–0.59 0.76–0.90

0.4 0.07 0.08 0.04 0.14

Band Widths (µm)

1379.46 1974.93 1743.12 1485.23 1041.28

Irradiance (W m−2 sr−1 µm−1)

10,000, 625

Data Points (# per hectares)

3–28

Frequency of Revisit (days)

a

Source: Adapted and edited from Melesse et al., 2007; Thenkabail P.S. et al. 2010. Journal Remote Sensing 2(1): 211–261. Of the 242 bands, 196 are unique and calibrated. These are: (A) Band 8 (427.55 nm) to band 57 (925.85 nm) that are acquired by visible and near-infrared (VNIR) sensor; and (B) Band 79 (932.72 nm) to band 224 (2395.53 nm) that are acquired by shortwave infrared (SWIR) sensor.

Spatial (meters)

1–4

Sensor

23. KOMPSAT-2

Radiometric (bit)

TABLE 1.1 (continued) Broad-Band and Narrow-Band Satellite Sensor Spatial, Spectral, Radiometric, Waveband, and Other Data Characteristics

10 Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

11

sensors such as Hyperion on-board the Earth Observing-1 (EO-1) satellite (Pearlman et al., 2003, Ungar et  al., 2003), and upcoming Hyperspectral Imaging Spectrometer and Infrared Imager (HypspIRI). These sensors gather data in 210–220 narrow-bands from 380 to 2500 nanometers at 60 m resolution or better (Table 1.1). The HyspIRI’s Thermal Infrared (TIR) sensor has 7 bands in the 7500–12,000 nm range that saturates at 400 K and 1 band in 3000–5000 nm (centered at 3980 nm) that saturates at 1400 K and acquires data in 60 m pixel size. However, it must be noted that using hyperspectral data are much more complex than using multispectral data. Hyperspectral systems collect large volumes of data in a short time leading to a number of issues that need to be addressed. For example, Hyperion, the first spaceborne hyperspectral sensor launched by NASA’s New Millennium Program (NMP) gathers near-continuous data in 220 discrete narrow-bands along the 400–2500 nanometer spectral range at a 30-meter spatial resolution and in 12-bit bytes with 12-bit radiometry. Each image is 7.5 kilometers in swath by 100 kilometers along track. The volume of data collected using Hyperion for an area equivalent to the Landsat TM image area will increase by about 37 times. Such increases in data volume pose great challenges in data handling. The issues include data storage volume, data storage rate, downlink or transmission bandwidth, real-time analog to digital bandwidth and resolution, computing bottle necks in data analysis, and new algorithms for data utilization (e.g., atmospheric correction is more complicated; Thenkabail et al., 2002, 2004b). These issues make it imperative that newer methods and techniques be developed to handle these highdimensional datasets. Future generations of satellites may carry specially optimized sensors to gather data for targeted applications. Or they may carry a narrow-waveband hyperspectral sensor like Hyperion and HyspIRI from which users with different application needs can extract appropriate optimal wavebands. However, having too many narrow-bands does not necessarily mean more information. Indeed, most of these bands, and especially the ones that are close to one another, provide redundant information. This will make users devote unnecessary time in data mining, complex processing to identify and remove redundant bands, and put a heavy burden on computing, processing, and storage resources. A far better option will be to focus on the design of an optimal sensor for a given application such as for vegetation studies and by dropping redundant bands. Even when the data are acquired in full range of hundreds or thousands of hyperspectral narrow-bands, knowledge of optimal bands for a particular application will help. Investigators can quickly select these bands, and spend time and expertise resources in applying selected optimal bands for the required application. Optimal hyperspectral sensors will help reduce data volumes, eliminate the problems of high-dimensionality of hyperspectral datasets, and make it feasible to apply traditional classification methods on a few selected bands (optimal bands) that capture most of the crop characteristics (Thenkabail et al., 2002, Thenkabail et  al., 2004a, Chan and Paelinckx, 2008, Dalponte et  al., 2009, Yang et  al., 2009). Thereby, knowledge of application specific “optimal bands” for high-dimensional datasets such as Hyperion and HyspIRI is mandatory to reduce costs in data analysis and computer resources. Table 1.1 compares the spectral and spatial resolution of narrow-band and broad-band data that are currently in use or are soon to be put into use. A number of recent studies have indicated the advantages of using discrete narrow-band data from a specific portion of the spectrum when compared with broad-band data to arrive at optimal quantitative or qualitative information on crop or vegetation characteristics (e.g., Thenkabail et al., 2000, Thenkabail et al., 2002, 2004a, Yang et al., 2009, Blackburn and Ferwerda, 2008, Chan and Paelinckx, 2008, Dalponte et al., 2009). Hence, this approach or direction has caught the attention of investigators and the community is moving in this direction. Given this background, the overarching goal is to explore and determine the optimal hyperspectral narrow-bands in the study of vegetation and agricultural crops and to enumerate on methods and approaches. This is partially to overcome the “curse” of high-dimensionality or Hughes phenomenon where the ratio of the number of pixels with known class identity (i.e., training pixels) and the number of bands must be maintained at or above minimum value to achieve statistical confidence

12

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

and functionality. In hyperspectral data with hundreds or even thousands of wavebands, the training pixels needed can grow exponentially (Hughes phenomenon), making it very difficult to address this spectral diversity. First, it is necessary to identify hyperspectral narrow-bands that are best suited for studying natural vegetation and agricultural croplands. In the process, one detects and eliminates redundant bands or examples that supply little knowledge to the application. This will lead to highlighting optimal hyperspectral wavebands, in the 400–2500 nanometer range, best suited to study vegetation and agricultural crops. There are a number of studies (e.g., Blackburn, 1998, Nolin and Dozier, 2000, Thenkabail et al., 2000, 2002, 2004a,b, Chan and Paelinckx, 2008, Dalponte et al., 2009, Yang et al., 2009, Yao et al., 2010) that indicate that the narrow wavebands located in specific portions of the spectrum have the capability to provide required optimal information sought for a given application. However, there is a clear need for synthesis of these various studies to see a general consensus of optimal wavebands based on studies conducted in different parts of the world for varying vegetation types and agricultural practices. Optimal Hyperion wavebands determined for vegetation studies established in this study will help reduce data volumes, eliminate the problems of high-dimensionality of hyperspectral datasets, and make it feasible to apply traditional classification methods on a few selected bands (optimal bands) that capture most of the information of vegetation and agricultural croplands. Secondly, it is important to present and discuss methods and approaches of hyperspectral data analysis. This is to quickly eliminate redundant bands and identify the most useful bands. For example, when a user receives hyperspectral data and their application is vegetation and agricultural studies, they can directly select wavebands recommended and ignore all other wavebands which are redundant. The process establishes categorization approaches to achieve highest accuracies, develop indices and wavebands that best model biophysical and biochemical quantities, and identify and eliminate redundant bands. The process involves an exhaustive review of the performance of hyperspectral vegetation indices (HVIs) that will help establish the utility of wavebands associated with these indices in the study of vegetation and agricultural crops.

1.2 HYPERSPECTRAL REMOTE SENSING OF VEGETATION AND AGRICULTURAL CROPS Agricultural crops are significantly better characterized, classified, modeled, and mapped using hyperspectral data. There are many studies supporting this, conducted on a wide array of crops and their biophysical and biochemical variables such as yield (Wang et al., 2008), chlorophyll a and b (Zhu et al., 2007, Delegido et al., 2010), total chlorophyll (Haboudane et al., 2004), nitrogen content (Rao et al., 2007), carotenoid pigments (Blackburn, 1998), plant stress (Zhao et al., 2005), plant moisture (Penuelas et al., 1995), above ground biomass (Shen et al., 2009), and biophysical variables (Darvishzadeh et al., 2008a,b,c). Hyperspectral remote sensing is well suited for early detection of nitrogen (N) as an important determinant of crop productivity and its quality. Knowing nitrogen levels early in the growing season will help remedy any deficiency. Leaf nitrogen is a key element in monitoring crop growth stage, its fertilization status, and assessing productivity. Leaf N accumulation, as a product of leaf N content and leaf weight, reflect not only information on leaf N status, but also vegetation coverage during crop growth (Yao et al., 2010). For example, nitrogen deficiency can be evaluated using ratios such as R743/R1316 (Abdel-Rahman et al., 2010), where “R” is radiance at the given narrowband in nanometers. Thenkabail et  al. (2000, 2002) studied numerous crops on marginal lands (see Figure 1.1a) that included cotton (Gossypium), potato (Solanum Erianthum), soybeans (Glycine max), corn (Zea mays), sunflower (Helianthus), barley (Hordeum vulgare L.), wheat (Triticum aestivum L. or Triticum durum Desf.), lentil (Lens esculenta Moench. or Lens orientale [Boiss.] Schmalh. or

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

13

(a)

(b)

FIGURE 1.1 Hyperspectral characteristics illustrated for few vegetation and agricultural crops. Hyperspectral narrow-band data obtained from ASD spectroradiometer illustrated for: (a) agricultural crops, (b) shrub species. (Continued)

Lens culinaris Medikus), cumin (Cuminum cyminum L.), chickpea (Cicer arietinum L.), and vetch (Vicia narbonensis L.). The parameters studied and found useful included: LAI or the leaf area index (m2/m2), WBM or wet biomass (kg/m 2), the above ground plant height (meters), CP or plant crude protein (%), N or nitrogen (%), and CC or mean canopy cover (%). Thousands of spectral measurements were made. These comprehensive studies led to identifying 12 optimal narrow-bands in 400–1050 nm (Thenkabail et al., 2000, 2002).

14

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation (c)

(d)

FIGURE 1.1 (Continued) Hyperspectral characteristics illustrated for few vegetation and agricultural crops. Hyperspectral narrow-band data obtained from ASD spectroradiometer illustrated for: (c) weed species. Hyperspectral narrow-band data obtained from Hyperion sensor onboard Earth Observing-1 (EO-1) satellite illustrated for: (d) tropical rainforest vegetation.

Forest and savanna biotic variables (e.g., height, basal area, biomass, and LAI) are modeled and mapped with significantly greater accuracies using hyperspectral data (Schlerf et al., 2005, Kalacska et al., 2007, Asner and Martin, 2009, Asner et al., 2009, Stagakis et al., 2010; e.g., Figure 1.1b–d) when compared to broad-band data. Ability to distinguish species type, assess age, structure (e.g., leaf angle), and phenology (e.g., leaf pigmentation, flowering), and map shade trees and understory crops is becoming increasingly possible as a result of hyperspectral narrow-band data (Oldeland et al., 2010, Papeş et al., 2010).

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

15

When there are significant topographic effects such as shadows cast from ridges, LiDAR data can be a great addition to hyperspectral data (Anderson et al., 2008). Thenkabail et al. (2004a,b) conducted a hyperspectral study of vegetation in African savannas and rainforests using Hyperion and an ASD spectroradiometer—both collecting data in the 400–2500 nm range (Table 1.1). They collected spectra from thousands of locations (see the few examples in Figure 1.1) that included an extensive and comprehensive series of vegetation types such as: (a) weeds (Imperata cylindrical the most prominent weed of African savannas and Chromolenea Odorata the most prominent weed in African rainforests), (b) shrubs (e.g., Pteleopsis habeensis, Piliostigma thonningii, Isoberlinia tomentosa; Figure 1.1b), (c) grasses (e.g., Brachiaria jubata, Brachiaria stigmatisata, Digitaria sp.), (d) agricultural crops (corn, rice, cowpea, groundnut, soybean, cassava), (e) other weeds (e.g., Ageraturm conyzoids, Aspilia Africana, Tephrosia bracteolate, Cassia obtusifolia), (f) agricultural fallows (e.g., of different age years), (g) primary forest (e.g., pristine, degraded), (h) secondary forest (e.g., young, mature, mixed), (i) regrowth forest (e.g., 8 years), (j) slash and burn agriculture, and (k) forest tree species (e.g., Piptadenia africana, Discoglypremna coloneura, Antrocaryon klaineanum, Pycnantus angolensis, Rauwolfia macrophylla, Alstonia congensis, Cissus spp., Lacospermas secundii, Haumania delkelmaniana, Alchornea floribunda, Lacospermas secundii, Alchornea cordifolia, Elaeis guineensis, Elaeis guineensis, Pteredium aquilinium, Megaphrenium spp., Pteredium aquilinium, Megaphrenium spp., Pteredium aquilinium, Epatorium adoratum, Aframomum giganteum, Pteredium acquilinium). The opportunities to distinguish various species using hyperspectral data are many. For example, the shrub species Isoberlinia tomentosa has high reflectivity in the red and NIR bands when compared with Piliostigma thonningii, but in the SWIR bands, it is just the opposite (Figure 1.1b). The shrub species Pteleopsis habeensis has dramatically high reflectivity relative to the other two shrub species (Piliostigma thonningii, Isoberlinia tomentosa) but beyond 2000 nm, Isoberlinia tomentosa has higher reflectivity than the other two species (Figure 1.1b). Similarly, the weed species Aspilia Africana has higher reflectivity relative to the other weed species (Ageraturm conyzoids, Tephrosia bracteolate, Cassia obtusifolia) only beyond 2000 nm. These results clearly imply the value of using data from distinct parts of the electromagnetic spectrum. These studies resulted in recommending 22–23 optimal bands to study vegetation in 400–2500 nm (Thenkabail et al., 2004a,b). The relationship between vegetation characteristics and spectral indices can vary considerably based on vegetation types, crop types, and species types (Papeş et al., 2010). Biomass and NDVI relationships within and between agricultural crops and various vegetation types are very distinct and are useful discriminators. These differences occur as a result of distinctive features such as (Thenkabail et al., 2002): (a) plant structure (e.g., erectophile and planophile, Figure 1.1a), (b) plant composition (e.g., nitrogen, lignin, chlorophyll a and b), and (c) quantitative characteristics (e.g., biomass per unit area, LAI, and plant height). Thereby, more specific and accurate attempts at biomass estimations using remotely sensed data of various characteristics will need to take into consideration the specific vegetation, land cover, crop type, and species types (Thenkabail et al., 2000). Specific advances are using radiative transfer models to derive quantities such as LAI and biomass of forests (e.g., Papeş et al., 2010, Schull et al., 2010). These models require surface reflectance and land-cover products. Accuracy of modeled quantities will depend on the number of land covers that can be mapped. Recent studies have also established the possibilities of direct biomass measurements using LiDAR that will provide within-canopy structure, ground topography, and tree height, thus enabling the indirect estimation of biomass and leaf area indices (Anderson et al., 2008). From the literature and hard work of investigators, it has been established that hyperspectral narrow-bands provide significant additional information when compared with similar information obtained from broad-band sensors in estimating biophysical characteristics (Wang et al., 2008, Asner and Martin, 2009), biochemical properties (Blackburn, 1998, Cho, 2007), forest classification (le Maire et al., 2008), and in other vegetation studies involving forests, grasses, shrubs, and weeds (Thenkabail et al., 2004a,b). This is due to many limitations of the broad-band data.

16

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

For example, most of the indices derived using the older generation of sensors tend to saturate (Thenkabail et al., 2000), resulting in significant limitations in distinguishing the species or modeling biophysical and biochemical quantities. Nondestructive measurements of vegetation characteristics such as canopy chemical content (e.g., chlorophyll, nitrogen), forest species, chemistry distribution, timber volumes, and plant moisture (Haboudane et al., 2002, 2004, Zarco-Tejada et al., 2005, Galvão et al., 2009), and biophysical and yield characteristics (Blackburn, 1998, Thenkabail et al., 2000, 2002, 2004a,b) are best measured using hyperspectral data. Hyperspectral narrow-band data have also the potential to separate vegetation into taxonomic levels (Ustin et al., 2004) and forest ecotypes (Chan and Paelinckx, 2008). But, this aspect still needs considerably more study. These efforts cover a wide array of agricultural crops and vegetation classes from forests, rangelands, savannas, and other ecosystems. They also cover a wide range of agro-ecologies and ecosystems spread across the world, resulting in a more holistic view of hyperspectral data performance in the study of agricultural crops across agro-ecological systems as well as forests and other vegetation across varied ecosystems (Hamilton, 2005, Gillespie et al., 2008). During this process, we also took into consideration that the high-dimensionality of hyperspectral data can, at times, lead to over-fitting of statistical models (Thenkabail et al., 2000) leading to an over-optimistic view of their power (Lee et al., 2004). This can be unfortunate, but once the issue is understood, approaches can be elucidated and employed to avoid difficulties. Taking all these factors into consideration, we set forth two key objectives, which were to: (a) establish optimal hyperspectral narrow-bands best suited to study vegetation and agricultural crops by pooling collective findings from a comprehensive set of investigations, and (b) identify and discuss a suite of methods and approaches that help to rapidly identify hyperspectral narrow-bands best suited to study through data mining of hundreds or thousands of wavebands. These objectives were met through the cumulative efforts of the co-editors and co-authors of the Second Edition,

1.3 HYPERSPECTRAL DATA COMPOSITION FOR STUDY OF VEGETATION AND AGRICULTURAL CROPS Hyperspectral narrow-bands, typically, contain 100–1500 wavebands and collect data in a nearcontinuous spectra from ultraviolet, visible, near-, mid-, and far-infrared; Figure 1.1. Resulting datasets offer many opportunities to study specific vegetation variables such as leaf area index (LAI) (Koger et al., 2003), biomass (Lee et al., 2004, Darvishzadeh et al., 2008a), vegetation fraction (Asner and Heidebrecht, 2002), canopy biochemistry (Asner and Heidebrecht, 2002), forest structure and characteristics (e.g., diameter at breast height, tree height, crown size, tree density) (Kalacska et al., 2007, White et al., 2010), and plant species (Martin et al., 1998, Oldeland et al., 2010). These data in hundreds or even thousands of bands are composited into a single hyperspectral data cube (HDC; Figure 1.2). For example, a click on any pixel of Hyperion HDC (see Figure 1.2), will provide a continuous spectrum describing the composition of that pixel (see Figure 1.1d). The principles of hyperspectral data composition and analysis can also be applied to time-series multispectral data. For example, a MODIS NDVI monthly maximum value composite for one year is composed as a single mega-file data cube (MFDC) (see Figure 1.3). Time-series spectra of a few classes extracted from this simple 12-band MFDC is illustrated in Figure 1.3, which is akin to hyperspectral data (Figure 1.1d) extracted from HDC (Figure 1.2).

1.4 METHODS AND APPROACHES OF HYPERSPECTRAL DATA ANALYSIS FOR VEGETATION AND AGRICULTURAL CROPS The goal of dimension reduction is to reduce the number of features substantially without sacrificing significant information. However, care should be taken to retain all key wavebands, so as not to: (a)

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

17

FIGURE 1.2 Hyperspectral data cube (HDC). The 242 band Hyperion image composed as HDC for 2 areas of African rainforests. Spectral signatures derived for a few classes from this Figure are illustrated in Figure 1.1d.

FIGURE 1.3 Mega-file data cube (MFDC). Spectral signatures extracted for a few classes from the 12 band MODIS monthly NDVI maximum value composite time-series MFDC. Akin to hyperspectral data (see Figures 1.1 and 1.2), this multispectral time-series is composed as a hyperspectral data cube. All hyperspectral data analysis techniques can be applied here. Note: illustrated for the country of Tajikistan.

18

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

reduce the discrimination power, (b) result in models that lead to lower accuracies, and (c) lead to models that fail to explain maximum variability.

1.4.1

Lambda (λ1) versus Lambda (λ2) PLots

Hyperspectral narrow-band data has high inter-band correlation leading to multiple measurements of the same quantity (Melendez-Pastor et al., 2010). Thereby, one of the key steps in hyperspectral data analysis is to identify and remove redundant bands. Using all the bands will only add burden to computing resources and does not add to additional information content. For example, wavebands that are close to one another (e.g., 680 nm vs. 690 nm), typically, are highly correlated (e.g., R2-square value of 0.99) and provide similar information. Thereby, using one of the two bands will suffice. Lambda versus Lambda plots determines (Figure 1.4): (a) redundant bands and (b) unique bands. Figure 1.4 shows inter-band correlation of rainforest vegetation gathered using 242 Hyperion bands (in 400–2500 nm with each band having 10 nm band-width) plotted as Lambda 1 (λ1 = 400–2500 nm) versus Lambda 2 (λ2 = 400–2500 nm). In Figure 1.4, the least redundant bands (R2-square values of 10 mimics linear mixing of soil and vegetation, as in linear mixture models. It is possible to compute SA HTBVIs for each of the 24,090 HTBVIs computed in Section 1.4.2. To reduce the computing time and optimize resources, SA HTBVIs can be computed only for the best HTBVIs (Figure 1.5). The value of computing SA HTBVIs is to enhance the sensitivity of HTBVIs by normalizing soil background influences; thus, helping to explain greater variability in crop biophysical and biochemical variables such as biomass, LAI, nitrogen, and chlorophyll a.

24

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

1.4.4.4.2 Atmospherically Resistant Hyperspectral Two-Band Vegetation Indices (AR HTBVIs) The atmospherically resistant vegetation index (ARVI) was first proposed by Kaufman and Tanre (1996) and was developed to account for aerosol effects by using the difference in blue and red reflectances to derive the surface red reflectance. This index is especially useful to study leaf pigment. Galvão et al. (2009) defined ARVI for Hyperion data as: ((λ864 − (2 * λ671 − λ467))/(λ864 + (2 * λ671 − λ467))

(1.5)

Another approach to minimize atmospheric effects on NDVI is to use the middle-infrared wavelength region (1.3–2.5 µm) as a substitute for the red band, since longer wavelengths are much less sensitive to smoke and aerosols (Jiang et al., 2008). Again, it is possible to compute AR HTBVIs for each of the 24,090 HTBVIs computed in Section 1.4.2. To reduce the computing time and optimize resources, AR HTBVIs can be computed only for the best HTBVIs (Figure 1.5). The value of computing AR HTBVIs is to enhance the sensitivity of HTBVIs by normalizing atmospheric aerosol and moisture effects; thus, helping to explain greater variability in crop biophysical and biochemical variables such as biomass, LAI, nitrogen, and chlorophyll a. 1.4.4.4.3 Hyperspectral Vegetation Indices of SWIR and TIR Bands (HVIST) Obtaining narrow-band data beyond 1100 nm, from the shortwave infrared (SWIR; 1100–2500 nm) and thermal infrared (TIR; 3000–14,000 nm), are important unique and/or complementing and/or supplementing information in addition to information on vegetation gained in the visible and near infrared (VNIR) discussed in previous paragraphs. The non-photosynthetic vegetation (NPV) such as litter, senesced leaves, and other dry vegetation are best differentiated based on lingo-cellulose bands in the SWIR (Numata et al., 2007, 2008) but are not spectrally separable from soil in the visible and near-infrared wavelength region (Asner and Lobell, 2000). Separating forest types and cropland classes, categorizing forest age classes or croplands at various stages of growth, modeling forest and cropland biotic factors such as canopy height, basal area, biomass, and LAI are often best predicted through a combination of visible and shortwave infrared (SWIR) (Cho, 2007, White et al., 2010). Moisture and plant stress properties are best quantified by including TIR bands along with SWIR, and VNIR (Kalacska et al., 2007, Galvão et al., 2009).

1.4.5

other methods of hyPersPectraL data anaLysis

There are numerous other methods of hyperspectral data analysis for studying vegetation and agricultural crop biophysical and biochemical properties. These include: A. Independent component analysis (Huang et al., 2018) which is an unsupervised temporal unmixing methodology that helps de-compose spectra. For example, it can recover both the time profile and area distribution of different crop types (Ozdogan, 2010); B. Wavelet transform can detect automatically the local energy variation of a spectrum in different spectral bands at each scale and provide useful information for hyperspectral image classification (Hsu, 2007); C. Radiative transfer models such as PROSPECT (le Maire et al., 2004) and LIBERTY (Coops and Stone, 2005) are used to simulate wide ranges of hyperspectral signatures and their impact on vegetation and crop variables; D. Minimum noise fraction (MNF) transformation separates noise from the data by using only the coherent portions, thus improving spectral processing results; and E. Spectral unmixing analysis (SMA; Shippert, 2001, Pacheco and McNairn, 2010) uses reference spectra (referred to as “endmembers”) in order to “unmix” characteristics of

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

25

spectra within each pixel. The biotic and abiotic characteristics of the constituents within a pixel will be gathered through ground truth data. The spectra of the individual constituents within the pixel can be obtained using a spectroradiometer and/or from spectral libraries. SMA classifications are known to map land cover more accurately than maximum likelihood classification (Okin et al., 2001, Pacheco and McNairn, 2010) and have the advantage of assessing the within pixel composition. Once all the materials in the image are identified, then it is possible to use linear spectral unmixing to find out how much of each material is in each pixel (Shippert, 2001).

1.4.6

broad-band vegetation index modeLs

It was determined by Thenkabail et al. (1999) that the vegetation indices computed for a wide range of existing broad-band sensors such as Landsat MSS, Landsat TM, SPOT HRV, NOAA AVHRR, and IRS-1C, using simulated data for crops from a spectroradiometer, were highly correlated (R2 value = 0.95 or higher). Hence, computing vegetation indices for any one of these sensors will provide nearly the same information as similar indices computed for other sensors. The most common categories of broad-band indices are: a. b. c. d.

1.5 1.5.1

NIR and red-based indices (Tucker, 1977); Soil-adjusted indices (Huete, 1988, 1989); Atmospheric resistant indices (Kaufman and Tanre, 1996); and Mid-infrared-based indices (Thenkabail et al., 1994a, 1995).

SEPARATING VEGETATION CLASSES AND AGRICULTURAL CROPS USING HYPERSPECTRAL NARROW-BAND DATA cLass seParabiLity using unique hyPersPectraL narrow-bands

There are fewer opportunities in separating vegetation classes using broad-bands from sensors such as Landsat TM (e.g., Figure 1.7a) when compared to numerous opportunities offered by narrowbands from sensors such as EO-1 Hyperion (e.g., Figure 1.7b). For example, the Landsat TM broadbands of near-infrared (TM4) versus red (TM3) could not separate a wheat crop from a barley crop (Figure 1.7a). However, when 2 narrow-bands involving the near-infrared centered at 910 nm with a narrow-band-width of 10 nm, versus red centered at 675 nm with narrow-band-width of 10 nm were used, an overwhelming proportion of the wheat crop fields were separated from barley crop fields (Figure 1.7b). Since hyperspectral sensors have 100 s of wavebands, the likelihood of finding wavebands that can separate vegetation/crop types (e.g., Figure 1.7b) or various groups of vegetation/ crop biophysical and biochemical quantities increases drastically. (a)

(b)

FIGURE 1.7 Separating 2 agricultural crops using broad-bands versus narrow-bands. The 2 broad-bands fail to separate wheat from barley whereas 2 distinct narrow-bands separate wheat from barley. The Landsat TM (a) and Hyperion (b) imaged the same surface at 30 m spatial resolution.

26

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

1.5.2

cLass seParabiLity using statisticaL methods

Stepwise discriminant analysis (SDA) is a powerful statistical tool for discriminating vegetation and agricultural crop types based on their quantitative and qualitative characteristics such as biophysical quantities, biochemical compositions, structural properties, and species type. The independent hyperspectral waveband variables are chosen to enter or leave the model using: (a) the significance level of the F test analysis of covariance, where the variables already chosen act as covariates and the variable under consideration is the dependent variable, or (b) the squared partial correlation for predicting the variable under consideration from the CLASS variable, controlling for the influences of the variables were selected for the model (SAS, 2010). Stepwise selection begins with no variable in the model. At each step, if a variable already in the model fails to meet the criterion to stay, the worst such variable is removed. Otherwise, the variable that contributes most to the discriminatory power of the model is entered. When all variables in the model meet the criterion to stay, and more of the other variables meet the criterion not to enter, the stepwise selection process stops (SAS, 2010). The stepwise discriminant analysis (Draper and Smith, 1981) was performed using the PROC STEPDISC algorithm of SAS (Klecka, 1980; SAS, 2010). Class separability in SDA can be expressed, most powerfully and lucidly, using: (a) Wilks’ Lambda, and (b) Pillai’s trace. Wilks’ Lambda and Pillai’s trace are based on the eigenvalues Γ of A*W−1 where A is the among the “sum-of-squares” (SS) and cross-products matrix, and W is the pooled SS and cross-products matrix: Wilks’ Λ = Π 1/(1 + λi) Pillai’s trace = Σ λi/(1 + λi)

(1.6)

Determinants (variance) of the S matrices are found. Wilks’ Lambda is the test statistic preferred for multivariate analysis of variance (MANOVA) and is found through a ratio of the determinants: Λ=

|Serror | |Seffect + Serror |

(1.7)

where, S is a matrix which is also known as: SS and cross-products, cross-products, or sum-ofproducts matrices. Wilks’ Lambda is the most commonly available and reported, however Pillai’s criterion is more robust and therefore more appropriate when there are small or unequal sample sizes. When separating 2 classes: A. The lower the value of Wilks’ Lambda, the greater is the separability between two classes. B. Higher the value of Pillai’s trace, greater is the separability between 2 classes. Class separability using hyperspectral data is illustrated for a few vegetation types in Figure 1.8. In each case, with the increase in number of wavebands, the separability also increases (Figure 1.8; note: lower the Wilks’ Lambda, greater the separability). For example, vegetation types and shrubs vs. grasses vs. weeds are best separated using around 17 hyperspectral narrow-bands (Figure 1.8), whereas, the 2 weeds (Chromolenea odorata vs. Imperata cylindrica) are separated best using 11 hyperspectral narrow-bands. It is possible to have a small incremental increase in spectral separability beyond 11 bands for weeds and beyond 17 bands for vegetation types, but those increases are very small and statistically insignificant. Figure 1.8 also shows that the fallows (1-year fallow

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

27

FIGURE 1.8 Class separability using hyperspectral narrow-bands determined based on Wilks’ Lambda. Lower the Wilks’ Lambda, greater the separability. So, with the addition of wavebands, the separability increases, reaching an optimal point beyond which the addition of wavebands does not make a difference.

vs. 1–3-year fallow) are not well separated by even hyperspectral narrow-bands since the Wilks’ Lambda remains high. In each case, it is possible to determine the exact wavebands that help increase separability (see Thenkabail et al., 2000, 2002, 2004a,b).

1.5.3

accuracy assessments of vegetation and croP cLassification using hyPersPectraL narrow-bands

Classification accuracies of vegetation and agricultural crops attained using hyperspectral narrow-band data are substantially higher than broad-bands (Bork et  al., 1999, Thenkabail et al., 2004b, Papeş et al., 2010, Schull et al., 2010). For example, greater than 90% classification accuracies have been obtained in classifying 5 agricultural crops using about 20 selected hyperspectral narrow-band data, relative to just about 60% accuracies obtained using 6 nonthermal broad-bands (e.g., Figure 1.9). Similarly, Dalponte et al. (2009) studied a forest area in Italy characterized by 23 different classes reaching accuracies of about 90% with hyperspectral data acquired at a spectral resolution of 4.6 nm in 126 bands. Clark et al. (2005) studied seven deciduous tree species with the HYDICE sensor, using three different classifiers, reaching accuracies to the order of 90%. Also, sensitivity to above ground biomass and ability to map forest successional stages are achieved using hyperspectral narrow-bands (Kalacska et al., 2007, Asner et al., 2009). Earlier, we established that selected hyperspectral narrow-bands can play a crucial role in separating 2 vegetation categories (e.g., Figure 1.7b) when 2 broad-bands fail to do so (e.g., Figure 1.7a). Also, the hyperspectral narrowbands explain about 10%–30% greater variability in modeling quantitative biophysical quantities (e.g., LAI, biomass) relative to broad-band data (Thenkabail et al., 2002, Chan and Paelinckx, 2008). The specific narrow-bands that contribute to increased accuracies were also identified.

28

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 1.9 Classification accuracies using hyperspectral narrow-bands versus Landsat broad-bands. The accuracies increased by 25%–30% when about 25 hyperspectral narrow-bands are used to classify 5 agricultural crops when compared to 6 Landsat broad-bands. Classification accuracies reach about 95% with 30 bands, beyond which accuracies do not increase. (Thenkabail P. S. et al. 2004b. Remote Sensing of Environment 91(2–3): 354–376.)

1.6 OPTIMAL HYPERSPECTRAL NARROW-BANDS IN THE STUDY OF VEGETATION AND AGRICULTURAL CROPS Optimal hyperspectral narrow-bands in the study of vegetation and agricultural croplands (Table 1.2) were determined based on an exhaustive review of literature that included: (a) identifying redundant bands (e.g. Figure 1.4), (b) modeling by linkage of crop biophysical and biochemical variables with hyperspectral indices and wavebands (e.g., Figure 1.5), (c) establishing wavebands that best help separate vegetation and crop types (e.g., Figure 1.7), (d) identifying wavebands, through statistical and other approaches, that best separate vegetation and crop types and characteristics (e.g., Figure 1.8), and (e) establishing classification accuracies of vegetation and crop classes and identifying wavebands that best help enhance these accuracies (e.g., Figure 1.9). The 28 wavebands listed in Table 1.2 are based on their frequency of occurrence in numerous studies discussed here. These prominent wavebands include: (a) two red bands centered at 675 nm (chlorophyll absorption maxima) and 682 nm (most sensitive to biophysical quantities and yield); (b) three NIR bands at 845 nm (mid-point of NIR “shoulder” that is most sensitive to biophysical quantities and yield), 915 nm (peak spectrum in NIR, useful for computing crop moisture sensitivity index with 975 nm); and 975 nm (moisture and biomass sensitive band). The NIR portion of the spectrum is highly sensitive to changes in biophysical quantities and plant structure. For example, the more horizontal, planophile structure plant leaf contributes significantly to greater reflectance in NIR and greater absorption in the red when compared with the more vertical erectophile structure (Thenkabail et al., 2000). The erectophile structure leads to significant slope changes in spectra in the region of 740 nm to 940 nm; (c) five green bands: 515 nm, 520 nm, 525 nm, 550 nm, and 575 nm. These bands are overwhelmingly sensitive to biochemical properties. Band λ = 550 nm is strongly correlated with total chlorophyll. The green band λ centered at 520 nm provides the most rapid positive change in reflectance per unit change in wavelength anywhere in the visible portion of the spectrum. The green band λ centered at 575 nm provides the most rapid negative change in reflectance per unit change in wavelength anywhere in the visible portion of the spectrum. Overall, the green bands are very sensitive to plant/ leaf nitrogen and pigment. There are dramatic shifts in crop-soil spectral behavior around 568 and 520 nm when there is change in pigment content and chloroplast structure for different crop types, growth stages, and growing conditions; and for (d) three red-edge bands at 700 and 720 nm (sensitive to vegetation stress), and 740 nm (sensitive to vegetation nitrogen content). The red-edge bands are

375 466 490

515 520

525

550

575

675

682

700

720

740

1 2 3

4 5

6

7

8

9

10

11

12

13

Waveband Waveband Number # Center nm

Nitrogen accumulation: Leaf nitrogen accumulation

Stress and chlorophyll: Nitrogen stress, crop stress, crop growth stage studies

Stress and chlorophyll: Nitrogen stress, crop stress, crop growth stage studies

D. Red-Edge Bands

Chlorophyll absorption maxima: greatest crop-soil contrast is around this band for most crops in most growing conditions. Strong correlations with chlorophyll a and b Biophysical quantities and yield: leaf area index, wet and dry biomass, plant height, grain yield, crop type, crop discrimination

C. Red Bands

Nitrogen: leaf nitrogen, wetland vegetation studies Pigment, biomass changes: positive change in reflectance per unit change in wavelength of this visible spectrum is maximum around this green waveband. Sensitive to pigment. Vegetation vigor, pigment, nitrogen: positive change in reflectance per unit change in wavelength in maximum as a result of vegetation vigor, pigment, and nitrogen. Chlorophyll: Total chlorophyll; Chlorophyll/carotenoid ratio, vegetation nutritional and fertility level; vegetation discrimination; vegetation classification. Vegetation vigor, pigment, and nitrogen: negative change in reflectance per unit change in wavelength is maximum as a result of sensitivity to vegetation vigor, pigment, and N.

B. Green Bands

fPAR, leaf water: fraction of photosynthetically active radiation (fPAR), leaf water content Chlorophyll: Chlorophyll a and b in vegetation Senescing and loss of chlorophyll\browning, ripening, crop yield: Sensitive to loss of chlorophyll, browning, ripening, senescing, and soil background effects.

A. Blue Bands

Importance in Vegetation and Agricultural Cropland Studies

Reference

Chan and Paelinck (2008), Thenkabail et al. (2004a,b), Jensen (2000) Maire et al. (2008), Thenkabail et al. (2002), Penuelas et al. (1995) Chan and Paelinck (2008), Thenkabail et al. (2004a,b) (Continued)

Chan and Paelinck (2008), Thenkabail et al. (2004a,b) Galvão et al. (2011), Thenkabail et al. (2004a,b)

Yang et al. (2009), Chan and Paelinck (2008), Thenkabail et al. (2002)

Thenkabail et al. (2004a,b), Curran (1989) Galvão, L.S. et al. (2011), Thenkabail et al. (2004a), Blackburn (1998, 1999) Curran (1989)

Thenkabail et al. (2000, 2002, 2004a,b) Thenkabail et al. (2000, 2002, 2004a,b) Thenkabail et al. (2004a,b)

TABLE 1.2 Optimal Hyperspectral Narrow-Bands Recommended in the Study of Vegetation and Agricultural Crops

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops 29

845

915 975

1100

1215

1245

1316 1445

1518

1725

2035 2173

2260

2295

2359

14

15 16

17

18

19

20 21

22

23

24 25

26

27

28

Waveband Waveband Number # Center nm

Thenkabail et al. (2002), Penuelas et al. (1995) Yao et al. (2010), Adam and Mutanga (2009), Müller et al. (2008)

Maire et al. (2008)

Reference

Cellulose, protein, nitrogen: sensitive to crop stress, lignin, and starch

Moisture and biomass: moisture absorption trough in far shortwave infrared (FSWIR). A point of most rapid change in slope of spectra based on vegetation vigor, biomass. Stress: sensitive to soil background and plant stress

Moisture and biomass: moisture absorption trough in far shortwave infrared (FSWIR) Protein, nitrogen

Nitrogen: leaf nitrogen content of crops Vegetation classification and discrimination: ecotype classification; plant moisture sensitivity. Moisture absorption trough in early shortwave infrared (ESWIR) Moisture and biomass: A point of most rapid rise in spectra with unit change in wavelength in SWIR. Sensitive to plant moisture. Lignin, biomass, starch, moisture: sensitive to lignin, biomass, starch. Discriminating crops and vegetation.

G. Shortwave Infrared (SWIR) Bands

Zhao et al. (2007), Thenkabail et al. (2004a,b), Gitelson and Merzlyak (1997) Yao et al. (2010), Maire et al. (2008), Vaiphasa et al. (2007), Thenkabail et al. (2002) Thenkabail et al. (2004b) Chan and Paelinck (2008), Thenkabail et al. (2004a,b) Thenkabail et al. (2002), Nichol et al. (2000), Elvidge and Chen (1995) Yao et al. (2010), Chan and Paelinck (2008), Thenkabail (2000) Galvão et al. (2009), Thenkabail et al. (2004a,b), Penuelas et al. (1995)

Curran (1989) Thenkabail et al. (2002)

Biophysical quantities: sensitive to biomass and leaf area index. A point of most rapid rise in spectra with unit Abdel-Rahman et al. (2010) change in wavelength in far near-infrared (FNIR) Moisture and biomass: A point of most rapid fall in spectra with unit change in wavelength in FNIR. Sensitive Yao et al. (2010), Becker et al. (2005) to plant moisture. Thenkabail et al. (2002), Nichol et al. (2000), Water sensitivity: water band index, leaf water, biomass. Reflectance peak in 1050–1300 nm. Elvidge and Chen (1995)

F. Far NIR (FNIR) Bands

Biophysical quantities and yield: leaf area index, wet and dry biomass, plant height, grain yield, crop type, crop discrimination, total chlorophyll Moisture, biomass, and protein: peak NIR reflectance. Useful for computing crop moisture sensitivity index. Moisture and biomass: Center of moisture sensitive “trough”; water band index, leaf water, biomass

E. Near-Infrared (NIR) Bands

Importance in Vegetation and Agricultural Cropland Studies

TABLE 1.2 (continued) Optimal Hyperspectral Narrow-Bands Recommended in the Study of Vegetation and Agricultural Crops

30 Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

31

especially sensitive to crop stress and changes in total chlorophyll and have potential to form a useful drought index. These are sensitive to senescing rates, chlorophyll changes, browning, ripening, carotenoids, and soil background influences; (e) as are three FNIR bands at 1100 nm (sensitive to biophysical quantities), 1215 nm (sensitive to moisture), and at 1245 nm (sensitive to water), and (f) nine SWIR bands (1316, 1445, 1518, 1725, 2035, 2173, 2260, 2295, and 2359 nm). These wavebands are overwhelmingly sensitive to moisture and biochemical properties; as are (g) three blue bands (375, 466, and 490 nm) that are especially sensitive to fraction of photosynthetically active radiation (fPAR), and senescing. Band λ = 490 nm is especially sensitive to carotenoids, leaf chlorophyll, and senescing conditions. A nominal bandwidth (Δλ) of 10 nm can be used for all wavebands. Too narrow a waveband will lead to lower signal to noise ratios. Keeping a band-width (Δλ) of 10 nm will ensure that optimal information on a particular feature is captured rather than average conditions captured in broader bands. In general, there is overwhelming evidence that the hyperspectral wavebands and indices involving specific narrow wavebands provide significantly better estimates on vegetation and crop biophysical quantities (e.g., LAI, biomass, plant height, canopy cover), biogeochemistry (e.g., lignin, chlorophyll a and b, photosynthesis, respiration, nutrient use (e.g., through LAI, and fPAR), canopy water content, and improved discrimination of species and land cover.

1.7 CONCLUSIONS An overwhelming proportion of the hyperspectral narrow-bands can be redundant. So, it will be extremely important to identify and remove the redundant bands from further analysis to ensure best use for resources and to overcome the curse of high-dimensionality or Hughes phenomenon. This led us to identify 28 optimal hyperspectral narrow-bands (Table 1.2) best suited in the study of vegetation and agricultural crops. The wavebands were identified based on their ability to: (a) best model biophysical and biochemical properties, (b) distinctly separate vegetation and crops based on their species type, structure, and composition, and (c) accurately classify vegetation and crop classes. Some caution is needed in eliminating redundant bands, since some of the redundant bands in one application may be useful in some other. Taking this into consideration, methods and approaches of data mining to quickly identify key wavebands important for a given application (and as a result eliminate redundant bands) are highlighted and these key wavebands are then used in more detail studies of that particular application. References cited here highlighted computation and use of various hyperspectral vegetation indices (HVIs) in studying and determining the most valuable wavebands (Table 1.2). The specific importance of each of the wavebands to biophysical and biochemical properties of vegetation and agricultural crops are also called out in Table 1.2. Typically, there is more than one waveband for each crop variable. For example, biophysical properties and yield are best modeled using wavebands centered at 682, 845, and 1100 nm. Similarly, moisture sensitivity in leaf/plants can be studied using several wavebands centered at 915, 975, 1215, 1518, 2035, and 2260 nm. The 28 wavebands (Table 1.2) will also allow us to compute 378 unique hyperspectral two-band vegetation indices (HTBVIs). Overall, there is clear evidence that the combination of the 28 identified bands (Table 1.2) and various HVIs computed from them will suffice to best characterize, classify, model, and map a wide array of vegetation and agricultural crop types and their biophysical and biochemical properties.

ACKNOWLEDGMENT The authors would like to thank Ms. Zhouting Wu of Northern Arizona University (NAU) for preparing Figure 1.3. Comments from reviewers were much appreciated.

32

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

REFERENCES Abdel-Rahman, E.M., Ahmed, F.B., and Van den Berg, M. 2010. Estimation of sugarcane leaf nitrogen concentration using in situ spectroscopy, International Journal of Applied Earth Observation and Geoinformation 12(1): S52–S57. Adam, E., and Mutanga, O. 2009. Spectral discrimination of papyrus vegetation (Cyperus papyrus L.) in swamp wetlands using field spectrometry, ISPRS Journal of Photogrammetry and Remote Sensing 64(6): 612–620. Anderson, J.E., Plourde, L.C., Martin, M.E., Braswell, B.H., Smith, M.-L., Dubayah, R.O., Hofton, M.A., and Blair, J.B. 2008. Integrating waveform LiDAR with hyperspectral imagery for inventory of a northern temperate forest, Remote Sensing of Environment 112: 1856–1870. Asner, G.P. 1998. Biophysical and biochemical sources of variability in canopy reflectance, Remote Sensing of Environment 64: 234–253. Asner, G.P., and Heidebrecht, K.B. 2002. Spectral unmixing of vegetation, soil and dry carbon cover in arid regions: Comparing multispectral and hyperspectral observations, International Journal of Remote Sensing 23: 3939–3958. Asner, G.P., and Lobell, D.B. 2000. A biogeophysical approach for automated SWIR unmixing of soils and vegetation, Remote Sensing of Environment 74: 99–112. Asner, G.P., and Martin, R.E. 2009. Airborne spectranomics: Mapping canopy chemical and taxonomic diversity in tropical forests, Frontiers in Ecology and the Environment 7: 269–276. Asner, G.P., Martin, R.E., Ford, A.J., Metcalfe, D.J., and Liddell, M.J. 2009. Leaf chemical and spectral diversity in Australian tropical forests, Ecological Applications 19: 236–253. Atzberger, C. 2004. Object-based retrieval of biophysical canopy variables using artificial neural nets and radiative transfer models, Remote Sensing of Environment 93: 53–67. Becker, B.L., Lusch, D.P., and Qi, J. 2005. Identifying optimal spectral bands from in situ measurements of Great Lakes coastal wetlands using second-derivative analysis, Remote Sensing of Environment 97(2): 238–248. Blackburn, G.A. 1998. Quantifying chlorophylls and carotenoids at leaf and canopy scales: An evaluation of some hyperspectral approaches, Remote Sensing of Environment 66: 273–285. Blackburn, G. 1999. Relationships between spectral reflectance and pigment concentrations in stacks of broadleaves, Remote Sensing of Environment 70: 224–237. Blackburn, J.A. 2006. Hyperspectral remote sensing of plant pigments, Journal of Experimental Botany 58(4): 855–867. Blackburn, A.G. and Ferwerda, J.G. 2008. Retrieval of chlorophyll concentration from leaf reflectance spectra using wavelet analysis. Remote Sensing of Environment 112(4):1614–1632. Bork, E.W., West, N.E., and Price, K.P. 1999. Calibration of broad-and narrow-band spectral variables for rangeland covers component quantification, International Journal of Remote Sensing 20: 3641–3662. Broge, N.H., and Leblanc, E. 2001. Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density, Remote Sensing of Environment 76(2): 156–172. Carter, G. 1998. Reflectance wavebands and indices for remote estimation of photosynthesis and stomatal conductance in pine canopies, Remote Sensing of Environment 63: 61–72. Chan, J.C., and Paelinckx, D. 2008. Evaluation of Random Forest and Adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery, Remote Sensing of Environment 112(6): 2999–3011. Chen, J., Wang, R., Wang, C. 2008. A multiresolution spectral angle-based hyperspectral classification method. International Journal of Remote Sensing 29(11): 3159 –3169. Cho, M.A. 2007. Hyperspectral remote sensing of biochemical and biophysical parameters: The derivative red-edge “double peak feature,” a nuisance or an opportunity? Ph.D. dissertation. p. 206. International Institute for Geo-information science and earth observation (ITC). Enschede, The Netherlands. Cho, M.A., Skidmore, A.K., and Sobhan, I. 2009. Mapping beech (Fagus sylvatica L.) forest structure with airborne hyperspectral imagery, International Journal of Applied Earth Observation and Geoinformation 11(3): 201–211. Clark, M.L., Roberts, D.A., and Clark, D.B. 2005. Hyperspectral discrimination of tropical rain forest tree species at leaf to crown scales, Remote Sensing of Environment 96: 375–398. Coops, N.C., and Stone, C. 2005. A comparison of field-based and modelled reflectance spectra from damaged Pinus radiata foliage, Australian Journal of Botany 53: 417–429.

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

33

Colombo, R., Meroni, M., Marchesi, A., Busetto, L., Rossini, M., Giardino, C., and Panigada, C. 2008. Estimation of leaf and canopy water content in poplar plantations by means of hyperspectral indices and inverse modeling. Remote Sensing of Environment 112(4): 1820–1834 Curran, P. 1989. Remote sensing of foliar chemistry, Remote Sensing of Environment 30: 271–278. Curran, P.J., Foody, G.M., Lucas, R.M., Honzak, M., and Grace, J. 1997. The carbon balance of tropical forests: From the local to the regional scale. In: van Gardingen, P.R., Foody, G.M. and Curran, P.J., Editors. Scaling-up from cell to landscape, Cambridge University Press, Cambridge, pp. 201–227. Dalponte, M., Bruzzone, L., Vescovo, L., and Gianelle, D. 2009. The role of spectral resolution and classifier complexity in the analysis of hyperspectral images of forest areas, Remote Sensing of Environment 113(11): 2345–2355. Darvishzadeh, R., Skidmore, A., Atzberger, C., and van Wieren, S. 2008a. Estimation of vegetation LAI from hyperspectral reflectance data: Effects of soil type and plant architecture, International Journal of Applied Earth Observation and Geoinformation 10: 358–373. Darvishzadeh, R., Skidmore, A., Schlerf, M., and Atzberger, C. 2008b. Inversion of a radiative transfer model for estimating vegetation LAI and chlorophyll in a heterogeneous grassland, Remote Sensing of Environment 112: 2592–2604. Darvishzadeh, R., Skidmore, A., Schlerf, M., Atzberger, C., Corsi, F., and Cho, M. 2008c. LAI and chlorophyll estimation for a heterogeneous grassland using hyperspectral measurements, ISPRS Journal of Photogrammetry and Remote Sensing 63: 409–426. Daughtry, C.S.T., and Walthall, C.L. 1998. Spectral discrimination of Cannabis sativa L. leaves and canopies, Remote Sensing of Environment 64(2): 192–201. De Foresta, H., and Michon, G. 1994. Agroforestry in Sumatra—where ecology meets economy, Agroforestry Today 6(4): 12–13. Delegido, J., Alonso, L., González, G., and Moreno, J. 2010. Estimating chlorophyll content of crops from hyperspectral data using a normalized area over reflectance curve (NAOC), International Journal of Applied Earth Observation and Geoinformation 12(3): 165–174. Drake, J.B., Dubayah, R.O., Clark, D.B., Knox, R.G., Blair, J.B., Hofton, M.A., Chazdon, R.L., Weishampel, J.F., and Prince, S. 2002. Estimation of tropical forest structural characteristics using large-footprint lidar, Remote Sensing of Environment 79(2–3): 305–319. Draper, N.R., and Smith, H. 1981. Applied regression analysis, Wiley, New York. Elvidge, C.D., and Chen, Z. 1995. Comparison of broad-band and narrow-band red and near-infrared vegetation indices, Remote Sensing of Environment 54: 38–48. Fava, F., Colombo, R., Bocchi, S., Meroni, M., Sitzia, M., Fois, N., and Zucca, C. 2009. Identification of hyperspectral vegetation indices for Mediterranean pasture characterization, International Journal of Applied Earth Observation and Geoinformation 11(4): 233–243. Ferwerda, J.G., Skidmore, A.K., and Mutanga, O. 2005. Nitrogen detection with hyperspectral normalized ratio indices across multiple plant species, International Journal of Remote Sensing 26(18): 4083–4095. Filippi, A.M., and Jensen, J.R. 2006. Fuzzy learning vector quantization for hyperspectral coastal vegetation classification, Remote Sensing of Environment 100(4): 512–530. Galvão, L.S., Epiphanio, J.C.N., Breunig, F.M., and Formaggio, A.R. 2011. Crop type discrimination using hyperspectral data. In: Thenkabail, P., Lyon, J., and Huerte, A., Editors. Hyperspectral Remote Sensing of Vegetation, Taylor and Francis. Chapter 18. Galvão, L. S., Roberts, D.A., Formaggio, A.R., Numata, I., and Breunig, F.M. 2009. View angle effects on the discrimination of soybean varieties and on the relationships between vegetation indices and yield using off-nadir Hyperion data, Remote Sensing of Environment 113(4): 846–856. Gillespie, T.W., Foody, G.M., Rocchini, D., Giorgi, A.P., and Saatchi, S. 2008. Measuring and modelling biodiversity from space, Progress in Physical Geography 32: 203–221. Gitelson, A.A., and Merzlyak, M.N. 1997. Remote estimation of chlorophyll content in higher plant leaves, International Journal of Remote Sensing 18(12): 2691–2697. Gong, P., Pu, R., and Miller, J. 1995. Coniferous forest leaf area index estimation along the Oregon transect using compact airborne spectrographic imager data, Photogrammetric Engineering and Remote Sensing 61: 1107–1117. Guisan, A., Edwards, T.C. Jr., and Hastie, T. 2002. Generalized linear and generalized additive models in studies of species distributions: Setting the scene, Ecological Modelling 157(2–3): 89–100. Guyot, G., and Baret, F. 1988. Utilisation de la haute résolution spectrale pour suivre l’état des couverts végétaux, Proceedings of the 4th International Colloquium on Spectral Signatures of Objects in Remote Sensing. ESA SP-287 Aussois, France, pp. 279–286.

34

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Haboudane, D., Miller, J.R., Pattey, E., Zarco-Tejada, P.J., and Strachan, I.B. 2004. Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture, Remote Sensing of Environment 90: 337–352. Haboudane, D., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J., and Dextraze, L. 2002. Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture, Remote Sensing of Environment 81: 416–426. Hamilton, A.J. 2005. Species diversity or biodiversity? Journal of Environmental Management 75: 89–92. Hill, M.J. 2004. Grazing agriculture: Managed pasture, grassland, and rangeland. In: Ustin, S.L. Editor. Manual of Remote Sensing, volume 4, Remote Sensing for Natural Resource Management and Environmental Monitoring, John Wiley & Sons, Hoboken, NJ, pp. 449–530. Hochberg, E.J., Roberts, D.A., Dennison, P.E., Hulley, G.C. 2015. Special issue on the Hyperspectral Infrared Imager (HyspIRI): Emerging science in terrestrial and aquatic ecology, radiation balance and hazards, Remote Sensing of Environment 167: 1–5. Horler, D.N.H., Dockray, M., and Barber, J. 1983. The red edge of plant leaf reflectance, International Journal of Remote Sensing 4(2): 273–288. Houborg, R., and E. Boegh, 2008. Mapping leaf chlorophyll and leaf area index using inverse and forward canopy reflectance modelling and SPOT reflectance data. Remote Sensing of Environment 112:186–202. Houborg, R., Anderson, Norman, J., Wilson, T., and Meyers, T. 2009. Intercomparison of a “bottom-up” and “top-down” modeling paradigm for estimating carbon and energy fluxes over a variety of vegetative regimes across the U.S., Agricultural and Forest Meteorology 149: 1875–1895. Hsu, P.-H. 2007. Feature extraction of hyperspectral images using wavelet and matching pursuit, ISPRS Journal of Photogrammetry and Remote Sensing 62(2): 78–92. Huang, Y., Chen, Z.-X., Yu, T., Huang, X.-Z., and Gu, X.-F. 2018. Agricultural remote sensing big data: Management and applications, Journal of Integrative Agriculture 17(9): 1915–1931, ISSN: 2095-3119, https://doi.org/10.1016/S2095-3119(17)61859-8, http://www.sciencedirect.com/science/article/pii/ S2095311917618598. Huete, A.R. 1988. A soil-adjusted vegetation index (SAVI), Remote Sensing of the Environment 25: 295–309. Huete, A.R. 1989. Soil influences in remotely sensed vegetation-canopy spectra. In: Asrar, G., Editor. Theory and applications of optical remote sensing, Wiley, New York, pp. 107–141. Ingram, J.C., Dawson, T.P., and Whittaker, R.J. 2005. Mapping tropical forest structure in southeastern Madagascar using remote sensing and artificial neural networks, Remote Sensing of Environment 94(4): 491–507. Jensen, J.R. 2000. Remote sensing of the environment: An earth resource perspective (Prentice Hall Series in Geographic Information Science), ISBN: 0134897331. Jiang, J., Huete, A.R., Didan, K., and Miura, T. 2008. Development of a two-band enhanced vegetation index without a blue band, Remote Sensing of Environment 112(10): 3833–3845. Jiang, Z., and Huete, A. R. 2010. Linearization of NDVI based on its relationship with vegetation fraction, Photogrammetric Engineering and Remote Sensing 76(8): 965–975. Jollineau, M.Y., and Howarth, P.J. 2008. Mapping an inland wetland complex using hyperspectral imagery. International Journal of Remote Sensing 29(12): 3609–3671. Kalacska, M., Sanchez-Azofeifa, G.A., Rivard, B., Caeilli, T., White, H.P., and Calvo-Alvarado, J.C. 2007. Ecological fingerprinting of ecosystem succession: Estimating secondary tropical dry forest structure and diversity using imaging spectroscopy, Remote Sensing of Environment 108: 82–96. Kaufman, Y.J., and Tanré, D. 1992. Atmospherically resistant vegetation index (ARVI) for EOS-MODIS, IEEE Transactions on Geoscience and Remote Sensing 30(2): 261–270. Kaufman, Y.J., and Tanre, D. 1996. Strategy for direct and indirect methods for correcting the aerosol effect on remote sensing from AVHRR to EOS-MODIS, Remote Sensing of Environment 55: 65–79. Klecka, W.R. 1980. Discriminant analysis. Quantitative applications in social sciences series, No. 19. SAGE Publications, Thousand Oaks, CA. Koger, C.J., Bruce, L.M., Shaw, D.R., and Reddy, K.N. 2003. Wavelet analysis of hyperspectral reflectance data for detecting pitted morning glory (Ipomoea lacunosa) in soybean (Glycine max), Remote Sensing of Environment 86(1): 108–119. Lee, K.S., Cohen, W.B., Kennedy, R.E., Maiersperger, T.K., and Gower, S.T. 2004. Hyperspectral versus multispectral data for estimating leaf area index in four different biomes, Remote Sensing of Environment 91(3–4): 508–520. Liu, Z.-Y., Wu, H.-F., and Huang, J.-F. 2010. Application of neural networks to discriminate fungal infection levels in rice panicles using hyperspectral reflectance and principal components analysis, Computers and Electronics in Agriculture 72(2): 99–106.

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

35

Lucas, K.L., and Carter, G.A. 2008. The use of hyperspectral remote sensing to assess vascular plant species richness on Horn Island, Mississippi, Remote Sensing of Environment 112(10): 3908–3915. Lunetta, R., and Lyon, J. 2008. Accuracy assessment of remote sensor and GIS. CRC Press, Boca Raton, FL. Lyon, J.G., Yuan, D., Lunetta, R.S., and Elvidge, C.D. 1998. A change detection experiment using vegetation indices, Photogrammetric Engineering and Remote Sensing 64: 143–150. le Maire, G., Francois, C., and Dufrene, E. 2004. Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements, Remote Sensing of Environment 89: 1–28. le Maire, G., François, C., Soudani, K., Berveiller, D., Pontailler, J.Y., Bréda, N., Genet, H., Davi, H., and Dufrêne, E. 2008. Calibration and validation of hyperspectral indices for the estimation of broadleaved forest leaf chlorophyll content, leaf mass per area, leaf area index and leaf canopy biomass, Remote Sensing of Environment 112: 3846–3864. Maire, G.L., François, C., Soudani, K., Berveiller, D., Pontailler, J.V., Bréda, N., Genet, H., Davi, H., and Dufrêne, E. 2008. Calibration and validation of hyperspectral indices for the estimation of broadleaved forest leaf chlorophyll content, leaf mass per area, leaf area index and leaf canopy biomass, Remote Sensing of Environment 112(10): 3846–3864. Martin, M., Newman, S., Aber, J., and Congalton, R. 1998. Determining forest species composition using high resolution remote sensing data, Remote Sensing of Environment 65: 249–254. Martínez, B., and Gilabert, M.A. 2009. Vegetation dynamics from NDVI time series analysis using the wavelet transform, Remote Sensing of Environment 113(9): 1823–1842. Melendez-Pastor, I., Navarro-Pedreño, J., Koch, M., and Gómez, I. 2010. Applying imaging spectroscopy techniques to map saline soils with ASTER images, Geoderma 158(1–2): 55–65. Melesse, A., Weng, Q., Thenkabail, P., and Senay, G. 2007. Remote sensing sensors and applications in environmental resources mapping and modelling, Sensors 7(12): 3209–3241. Muller, K., Bottcher, U., Meyer-Schatz, F., and Kage, H. 2008. Analysis of vegetation indices derived from hyperspectral reflection measurements for estimating crop canopy parameters of oilseed rape (Brassica napus L.), Biosystems Engineering 101: 172–182. Mutanga, O., and Skidmore, A.K. 2004. Narrow-band vegetation indices overcome the saturation problem in biomass estimation, International Journal of Remote Sensing 25(2004): 1–16. Nichol, C., Huemmrich, K., Black, T., Jarvis, P., Walthall, C., Grace, J., and Hall, F. 2000. Remote sensing of photosynthetic light-use efficiency of boreal forest, Agricultural and Forest Meteorology 101: 131–142. Nolin, A.W., and Dozier, J. 2000. A hyperspectral method for remotely sensing the grain size of snow, Remote Sensing of Environment 74(2): 207–216. Numata, I., Roberts, D.A., Chadwick, O.A., Schimel, J.P., Galvão, L.S., and Soares, J.V. 2008. Evaluation of hyperspectral data for pasture estimate in the Brazilian Amazon using field and imaging spectrometers, Remote Sensing of Environment 112(4): 1569–1683. Numata, I., Roberts, D.A., Chadwick, O.A., Schimel, J., Sampaio, F.R., Leonidas, F.C., and Soares, J.V. 2007. Characterization of pasture biophysical properties and the impact of grazing intensity using remotely sensed data, Remote Sensing of Environment 109(2007): 314–327. Okin, G.S., Roberts, D.A., Murray, B., and Okin, W.J. 2001. Practical limits on hyperspectral vegetation discrimination in arid and semiarid environments, Remote Sensing of Environment 77: 212–225. Oldeland, J., Wesuls, D., Rocchini, D., Schmidt, M., and Jürgens, N. 2010. Does using species abundance data improve estimates of species diversity from remotely sensed spectral heterogeneity? Ecological Indicators 10: 390–396. Ozdogan, M. 2010. The spatial distribution of crop types from MODIS data: Temporal unmixing using independent component analysis, Remote Sensing of Environment 114(6): 1190–1204. Pacheco, A., and McNairn, H. 2010. Evaluating multispectral remote sensing and spectral unmixing analysis for crop residue mapping, Remote Sensing of Environment 114(10): 2219–2228. Papeş, M., Tupayachi, R., Martínez, P., Peterson, A.T., and Powell, G.V.N. 2010. Using hyperspectral satellite imagery for regional inventories: A test with tropical emergent trees in the Amazon Basin, Journal of Vegetation Science 21(2): 342–354. Pearlman, J.S., Barry, P.S., Segal, C.C., Shepanski, J., Beiso, D., and Carman, S.L. 2003. Hyperion, a spacebased imaging spectrometer, IEEE Transactions on Geoscience and Remote Sensing 41: 1160–1173. Penuelas, J., Filella, I., Lloret, P., Munoz, F., and Vilajeliu, M. 1995. Reflectance assessment of mite effects on apple trees, International Journal of Remote Sensing 16: 2727–2733. Rao, N.R., Garg, P.K., and Ghosh, S.K. 2007. Estimation of plant chlorophyll and nitrogen concentration of agricultural crops using EO-1 Hyperion hyperspectral imagery, Journal of Agricultural Science 146: 1–11.

36

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Sakamoto, T., Yokozawa, M., Toritani, H., Shibayama, M., Ishitsuka, N., and Ohno, H. 2005. A crop phenology detection method using time-series MODIS data, Remote Sensing of Environment 96: 366–374. SAS Institute Inc. 2010. SAS/STAT user’s guide, Version 6. 4th Ed., Volume 1, Cary, North Carolina. Schlerf, M., Atzberger, C., and Hill, J. 2005. Remote sensing of forest biophysical variables using HyMap imaging spectrometer data, Remote Sensing of Environment 95(2): 177–194. Schmidt, K.S., and Skidmore, A.K. 2003. Spectral discrimination of vegetation types in a coastal wetland, Remote Sensing of Environment 85(1): 92–108. Schull, M.A., Knyazikhin, Y., Xu, L., Samanta, A., Carmona, P.L., Lepine, L., Jenkins, J.P., Ganguly, S., and Myneni, R.B. 2010. Canopy spectral invariants, Part 2: Application to classification of forest types from hyperspectral data, Journal of Quantitative Spectroscopy and Radiative Transfer 112(4): 736–750. Shen, M., Tang, Y., Klein, J., Zhang, P., Gu, S., Shimono, A., and Chen, J. 2009. Estimation of aboveground biomass using in situ hyperspectral measurements in five major grassland ecosystems on the Tibetan Plateau, Journal of Plant Ecology 2(1): 43. Shippert, P. 2001. Spectral and hyperspectral analysis with ENVI, ENVI User’s Group Notes, April 22–28, Annual meeting of the American Society of Photogrammetry and Remote Sensing, St. Louis, MO. Stagakis, S., Markos, N., Sykioti, O., and Kyparissis, A. 2010. Monitoring canopy biophysical and biochemical parameters in ecosystem scale using satellite hyperspectral imagery: An application on a Phlomis fruticosa Mediterranean ecosystem using multiangular CHRIS/PROBA observations, Remote Sensing of Environment 114(5): 977–994. Thenkabail, P.S. 2003. Biophysical and yield information for precision farming from near-real time and historical Landsat TM images, International Journal of Remote Sensing 24(14): 2879–2904. Thenkabail, P.S., Smith, R.B., and De-Pauw, E. 1999. Hyperspectral vegetation indices for determining agricultural crop characteristics, CEO research publication series No. 1, Center for earth Observation, Yale University, New Haven, pp. 47. ISBN: 0-9671303-0-1. Thenkabail, P.S., Ward, A.D., and Lyon, J.G. 1995. Impacts of agricultural management practices on soybean and corn crops evident in ground-truth data and thematic Mapper vegetation indices, Transactions of the American Society of Agricultural Engineers 37(3): 989–995. Thenkabail, P., Lyon, G.J., Turral, H., and Biradar, C.M. 2009a. Remote Sensing of Global Croplands for Food Security, CRC Press-Taylor and Francis Group. Thenkabail, P.S., Biradar C.M., Noojipady, P., Dheeravath, V., Li, Y.J., Velpuri, M., Gumma, M., Reddy, G.P.O., Turral, H., Cai, X.L., Vithanage, J., Schull, M., and Dutta, R. 2009b. Global irrigated area map (GIAM), derived from remote sensing, for the end of the last millennium, International Journal of Remote Sensing 30(14): 3679–3733. July, 20, 2009. Thenkabail, P.S., Enclona, E.A., Ashton, M.S., Legg, C., and Jean De Dieu, M. 2004a. Hyperion, IKONOS, ALI, and ETM+ sensors in the study of African rainforests, Remote Sensing of Environment 90: 23–43. Thenkabail, P.S., Enclona, E.A., Ashton, M.S., and Van Der Meer, V. 2004b. Accuracy assessments of hyperspectral waveband performance for vegetation analysis applications, Remote Sensing of Environment 91(2–3): 354–376. Thenkabail, P.S., GangadharaRao, P., Biggs, T., Krishna, M., and Turral, H. 2007. Spectral matching techniques to determine historical land use/land cover (LULC) and irrigated areas using time-series AVHRR pathfinder datasets in the Krishna river basin, India, Photogrammetric Engineering and Remote Sensing 73(9): 1029–1040. Thenkabail, P.S., Hanjra, M.A., Dheeravath, V., and Gumma, M.A. 2010. A holistic view of global croplands and their water use for ensuring global food security in the 21st century through advanced remote sensing and non-remote sensing approaches, Journal Remote Sensing 2(1): 211–261. Thenkabail, P.S., Smith, R.B., and De-Pauw, E. 2000. Hyperspectral vegetation indices for determining agricultural crop characteristics, Remote Sensing of Environment 71: 158–182. Thenkabail, P.S., Smith, R.B., and De-Pauw, E. 2002. Evaluation of narrowband and broadband vegetation indices for determining optimal hyperspectral wavebands for agricultural crop characterization, Photogrammetric Engineering and Remote Sensing 68(6): 607–621. Thenkabail, P.S., Ward, A.D., and Lyon, J.G. 1994a. LANDSAT-5 Thematic Mapper models of soybean and corn crop characteristics, International Journal of Remote Sensing 15(1): 49–61. Thenkabail, P.S., Ward, A.D., Lyon, J.G., and Merry, C.J. 1994b. Thematic Mapper vegetation indices for determining soybean and corn crop growth parameters, Photogrammetric Engineering and Remote Sensing 60(4): 437–442. Trombetti, M., Riaño, D., Rubio, M.A., Cheng, Y.B., and Ustin, S.L. 2008. Multi-temporal vegetation canopy water content retrieval and interpretation using artificial neural networks for the continental USA, Remote Sensing of Environment 112(1): 203–215.

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

37

Tsai, F., Lin, E., and Yoshino, K. 2007. Spectrally segmented principal component analysis of hyperspectral imagery for mapping invasive plant species, International Journal of Remote Sensing 28: 1023–1039. Tucker, C.J. 1977. Spectral estimation of grass canopy variables, Remote Sensing of Environment 6: 11–26. Ungar, S.G., Pearlman, J.S., Mendenhall, J.A., and Reuter, D. 2003. Overview of the earth observing one (EO-1) mission, IEEE Transactions on Geoscience and Remote Sensing 41: 1149–1159. Ustin, S.L., Roberts, D.A., Gamon, J.A., Asner, G.P., and Green, R.O. 2004. Using imaging spectroscopy to study ecosystem processes and properties, Bio Science 54(2004): 523–534. Vaiphasa, C.K., Skidmore, K.A., de Boer, W.F., and Vaiphasa, T. 2007. A hyperspectral band selector for plant species discrimination, ISPRS Journal of Photogrammetry and Remote Sensing 62(3): 225–235. Wang, F.M., Huang, J.F., and Wang, X.Z. 2008. Identification of optimal hyperspectral bands for estimation of rice biophysical parameters, Journal of Integrative Plant Biology 50(3): 291–299. White, J.C., Gómez, C., Wulder, M.A., and Coops, N.C. 2010. Characterizing temperate forest structural and spectral diversity with Hyperion EO-1 data, Remote Sensing of Environment 114(7): 1576–1589. Yang, F., Li, J., Gan, X., Qian, Y., Wu, X., and Yang, Q. 2009. Assessing nutritional status of Festuca arundinacea by monitoring photosynthetic pigments from hyperspectral data, Computers Electronics in Agriculture 70(1): 52–59. Yao, X., Zhu, Y., Tian, Y.C., Feng, W., and Cao, W.X. 2010. Exploring hyperspectral bands and estimation indices for leaf nitrogen accumulation in wheat, International Journal of Applied Earth Observation and Geoinformation 12(2): 89–100. Zarco-Tejada, P.J., Berjón, A., López-Lozano, R., Miller, J.R., Martín, P., Cachorro, V., González, M.R., and de Frutos, A. 2005. Assessing vineyard condition with hyperspectral indices: Leaf and canopy reflectance simulation in a row-structured discontinuous canopy, Remote Sensing of Environment 99: 271–287. Zhao, D., Huang, L., Li, J., and Qi, J. 2007. A comparative analysis of broadband and narrowband derived vegetation indices in predicting LAI and CCD of a cotton canopy, ISPRS Journal of Photogrammetry and Remote Sensing 62(1): 25–33. Zhao, D.H., Li, J.L., and Qi, J.H. 2005. Identification of red and NIR spectral regions and vegetative indices for discrimination of cotton nitrogen stress and growth stage, Computers and Electronics in Agriculture 48(2): 155–169. Zhao, G., and Maclean, L. 2000. A comparison of canonical discriminant analysis and principal component analysis for spectral transformation, Photogrammetric Engineering and Remote Sensing 66: 841–847. Zhou, D., Zhu, Y., Tian, Y., Yao, X., and Cao, W. 2006. Monitoring leaf nitrogen accumulation with canopy spectral reflectance in rice, Acta Agronomica Sinica 32(9): 1316–1322. Zhu, Y., Li, Y., Feng, W., Yao, X., and Cao, W. 2006. Monitoring leaf nitrogen in wheat using canopy reflectance spectra, Canadian Journal of Plant Science 86: 1037–1046. Zhu, Y., Zhou, D., Yao, X., Tian, Y., and Cao, W. 2007. Quantitative relationships of leaf nitrogen status to canopy spectral reflectance in rice, Australian Journal of Agricultural Research 58(11): 1077–1085.

Section II Hyperspectral Sensor Systems

2

Hyperspectral Sensor Characteristics Airborne, Spaceborne, Hand-Held, and Truck-Mounted; Integration of Hyperspectral Data with LiDAR Fred Ortenberg

CONTENTS 2.1 2.2 2.3 2.4

Introduction ............................................................................................................................ 41 Hyperspectral Sensor Concept ...............................................................................................44 Hyperspectral Sensor Physics, Principles, and Design........................................................... 47 Hyperspectral Sensor Operational Modes .............................................................................. 50 2.4.1 Ground-Based Hyperspectral Imaging ....................................................................... 50 2.4.2 Airborne Hyperspectral Sensors ................................................................................ 54 2.4.3 Spaceborne Imaging ................................................................................................... 57 2.5 LiDAR and Hyperspectral Data Integration........................................................................... 62 2.6 Summary and Outlook ...........................................................................................................66 References ........................................................................................................................................ 67

2.1

INTRODUCTION

This chapter sets forth the fundamental concepts of hyperspectral imaging—a powerful tool to collect precise remote sensing data involved in detection, recognition, and examination of the different objects of scientific, economic, and defense character. The development of the relevant equipment and systems is also discussed, as well as a few generic applications. A hyperspectral sensor is one of the devices in use in remote sensing activities aimed at observing the Earth’s surface with ground-based, airborne, or spaceborne imaging gear [1–7]. In outline definition, remote sensing consists in photo or electronic recording of the spatial and spectral distributions of electromagnetic radiation, as emitted by any object under monitoring either on the Earth’s surface, atmosphere or underground. Present-day technologies enable such a recording throughout the entire spectrum, from the radio wavebands to the x-ray and gamma regions. The respective receivers differ widely, for each of them is based on specific physical principles optimized for collection in a given spectral region, zone or band. Basically, three distinct spectral regions are used in the remote sensing community, namely: visible light with very-near infrared (IR), thermal, and radio wavelengths. In each of them, an image of any object is unique; it is defined by the object’s specifics as to emission, reflection, absorption, and scattering of the electromagnetic waves. In the visible and very-near IR regions, it is the object’s capacity to reflect solar energy in the first place, determined by the chemical composition of its surface. The thermal IR region is characterized by radiated energy, the latter being directly 41

42

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

dependent on the object temperature. In the UHF radio region—that of the radars—the reflection is determined by the surface smoothness and texture. The sensor which operates in a visible waveband, can either cover the entire spectral region in a so-called panchromatic imaging, or be confined to some of the spectrum zones or bands (e.g., green or red). In the first case, the basic sensor’s characteristic is its resolution. At present, the sensors are developed of a very-high-resolution (VHR) class. A camera or instrument of such a class onboard a Low Earth Orbit (LEO) satellite allows Earth’s surface imaging with a resolution of less than 1 m. In the second case, a multi-zonal or multiband image yields a set of distinct zone-specific images, which can be directly synthesized into color variants to bring out the details of interest, such as roads, installations, water surfaces, and vegetation. The development of such equipment does not consist any more in enhancement of the images’ spatial resolution, but in the increase of the number of the images taken in different spectral zones, that is, of the number of spectral channels, at a reasonable spatial resolution of each image. The more zones or bands, the more useful information. With several zones, the sensor becomes multispectral. With many narrow spectral zones, the sensor becomes hyperspectral. It should be borne in mind, however, that the spectral region accessible to the human eye is but a minor source of such information. The ability of bees to single out melliferous herbs in a multi-hued summer meadow relates to the fact their vision extends into the ultra violet (UV) region. With such a “super-vision” directed at the Earth’s surface, it is possible to distinguish healthy vegetation from degraded vegetation and human-made objects from natural. Thermal IR radiation also carries particular information unattainable with the human eye. The natural radiation of any object indicates its surface temperature: the higher the latter, the lighter the shade of grey in the image. The eye being more sensitive to variations in hue than to those in brightness, the image contrasts are often deepened by means of a “scale” of the “false” colors commonly associated with temperatures—from violet, or blue (“cold”) to red, or brown (“hot”). Thermal photography or videography is a widely-used tool for detection of the heat leaks in both industrial installations and built-up residential areas. They are also useful in reconnaissance: an underground defense plant can be invisible in the ordinary photograph, but since its functioning entails heat generation, it would shine brightly in the thermal region. Finally, such a picture of a person can provide valuable information on their state of health; much like thermal IR thermometers used in clinics that do not “touch” the patient but rather remotely sense their skin temperature. Regarding the very-near IR region, it is possible to take any of its zones and substitute it for one of the primary colors in an ordinary color image. Such an operation, known as color synthesis, is performed by the “false-color” films used in remote sensing. The inexperienced eye may find it hard to recognize familiar objects in synthesized images but, on the other hand, they help overcome the limitations of human vision. The most promising method of remote sensing is hyperspectral or HS imaging, which produces tens or even hundreds of images in narrow zones. The absorption spectra of substances and materials being specific, this technique makes it possible to identify vegetation, minerals, geological formations, soils, and structural materials in buildings and pavements via the physical/chemical compositions of the objects. Thanks to the extra-high resolution (of the order of that of a lab spectrometer) the volume of obtainable information can be increased by a four-figure factor or more. Unlike the visible- and IR-region imaging devices, the radars belong to the active-sensor class. Whereas the former just passively capture either reflected and scattered solar radiation or that of the Earth’s surface, the latter instrument emits their own electromagnetic waves and records their echo off the object. Radar imaging has features absent in other remote sensing techniques: it can’t be hampered by blanket cloud cover; its images have their particular geometric distortions; and its coherence makes it possible to obtain detailed relief patterns accurate to tens of centimeters and it can operate both in daytime and at night. Hyperspectral imaging is an evolutionary product of the multispectral (MS) systems, whereby— by means of new technologies—the number of information collection channels can be increased

Hyperspectral Sensor Characteristics

43

from 3–10 to 100–1000 with a high spectral resolution of 1–10 nm. The result is a multidimensional spatial-spectral image, in which each picture element (pixel) is characterized by its individual spectrum. Such an image is called an information cube, with two of its dimensions representing the projection of the imaged area on the plane, and the third—the frequency of the received radiation. A hyperspectral imaging device is an opto-electronic multichannel system designed for simultaneous independent generation of an image and corresponding video signal in a discrete or continuous sequence of spectrum intervals, supplemented by radiometric, spectral, and spatial image parameters pertaining to those ranges. In other words, present-day multispectral and hyperspectral scanning devices are radiometrically calibrated multichannel video-spectrometers. The values of brightness, registered by the imaging system for some object in the different spectral zones, together with their graphical mappings serve as spectral curves allowing one to clearly discern the object and mark it off on the image. The term “hyperspectral” has been repeatedly criticized, for such a collocation seems to be incongruous, without any physical sense. More justified would be using the prefix “hyper-” with a quantitative characterization to imply “too many,” as in the word “hypersonic,” that is, highly above the speed of sound. But in our case “spectrum” has a physico-mathematical, rather than a quantitative meaning. Similarly, this is why the words like “hyperoptical,” “hyperoily,” “hypermechanical,” “hyperaerial,” and so on, are devoid of sense. Consequently, the word “hyperspectral” might represent a metaphor, to express the excitement of the researchers as regards to the new technological means able to obtain a multitude of images of the same object, each of them in a very narrow spectral band. Spectroradiometer imaging technology emerged at the time when the panchromatic imagery was sufficiently advanced, and the picture quality was defined by the achieved spatial resolution. The images taken from the satellite orbit were said to be either of “high resolution” (about 1 m), or of “very high resolution” (less than 1 m). By the term “resolution,” the spatial one was meant by default. Since the image sets have been provided by the spectroradiometer, the technique of obtaining these data cubes began to be referred to as the “hyper-resolution imaging.” Until now, some researchers keep using this term, even without specifying what resolution they mean—spatial or spectral. The expression “spectral hyper-resolution imaging” could have a meaning of the visual spectroradiometer data, but the specialists have chosen instead this derivation—“hyperspectral.” In any case, the term has taken root. Still, it should be made clear that what matters is not the discovery of a new phenomenon, but the development of the advanced video-spectrometers with high spectral resolution. The interest in such opto-electronic devices is caused not only by their spectacular applications, but also by the fact that these technical means in many respects imitate the elements of the visual apparatus of higher animals and humans. It’s known that all the living beings acquire 80%–90% of the information about their environment through vision. Therefore, the paths of developing the automatic systems of the technical vision for detection, recognition, and classification of the different objects have received close attention. Hyperspectral imaging, being able to extract more precise and detailed information as compared with other techniques, is one of the most efficient and fast-developing directions of Earth remote sensing. The data on the energy reflected from objects provides extensive material for detailed analysis. Still, it is worth mentioning that development of the needed hyperspectral equipment proved to be a very complex task and hard to implement. This is why for a long enough time, designing hyperspectral sensors, in particular for spaceborne applications, had been of purely theoretical interest for the following reasons: • The quality of the equipment precluded realization of hyperspectral sensors with high spatial and spectral resolution simultaneously. • Reception of the vast volumes of hyperspectral data, their transmission through the communication channels, and their ground-based processing presented serious difficulties. • The complexity of the data-flow containing the hyperspectral measurements required specialized and highly sophisticated processing software.

44

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Nowadays, these difficulties are largely surmountable. At the time of writing, ways are opened for development of hyperspectral equipment in the optical and IR regions with high informational and operational indices, which in turn would mean a higher solution level for the problems to be solved, and spaceborne sensors are being developed with an eye towards operational and global data collection. During the last three decades, the hi-tech industries in many countries were considerably involved in hyperspectral subjects. The dedicated element base was created for manufacturing the hyperspectral equipment. Various hyperspectral sensors, initially for the ground and airborne applications, and later on for the spaceborne ones, were designed, tested, and put to use. The modern spectrometer concept is based on the latest optical-system solutions involving large-format charge-coupled device (CCD) matrices. Currently available are imaging systems operating in hundreds of spectral bands, with signal-to-noise ratios providing information in 12 bits per count or byte. The advances in hyperspectral equipment and data processing software in recent years has significantly extended the range of possible applications, to be provided by the airborne and spaceborne imaging community. The specifics just mentioned of putting hyperspectral sensors into space resulted in smaller availability of hyperspectral imagery as compared with other Earth remote sensing data. One of the reasons was that of a small number of the spacecraft with hyperspectral imaging systems, like Hyperion onboard the Earth Observing-1 (EO-1) spacecraft (NASA, USA), or CHRIS onboard PROBA (ESA). The increased hyperspectral-connected activity in space can be seen as of late, to overcome the backlog. Spaceborne hyperspectral sensors are expected to find use as general purpose instruments, providing data for a broad range of end-users. Present-day airborne and spaceborne hyperspectral instrumentation operates in the optical and IR spectral bands and combines high spatial, spectral, and radiometric resolutions, and hence are now indispensable in recording distinctive features of the Earth’s surface such as vegetation cover, landscape status, and anthropogenic impacts. Its broad, functional potential derives from the fact that the object’s spectral parameters, as well as their surface distributions, are the most prolific sources of information on the object’s state and the changes it is undergoing in the course of exploitation or its vital activity. For the hyperspectral imaging techniques evolving and getting introduced into new fields, their primary contribution will be in the exploration and development of the new applications via the selection of optimal spectral band parameters (bands position and widths). Practical operational considerations (sensor cost, data volume, data processing costs, etc.) for most current and future applications are favorable for the use of hyperspectral systems, taking into account their economic efficiency. Aerial and spaceborne hyperspectral sensors potentially provide an economic alternative to traditional ground-based observations alone, permitting data to be acquired for large areas in a minimum of time while enabling faster delivery and application turnaround. By virtue of its ability to provide information beyond the capacity of traditional panchromatic and multispectral photography and imaging, hyperspectral sensor data is highly advantageous in solving economic, scientific and engineering problems. Some experts estimate [8] that hyperspectral sensors are capable of solving up to 70% of all Earth-observation problems, while visual information with high spatial resolution—only 30%. As will be shown later in this and other chapters, additional information about the reflecting surfaces can be delivered by the return signals measured by LiDAR systems. This is why some new hyperspectral applications are based on the fusion of two imaging data types, as obtained by joint airborne or spaceborne hyperspectral and LiDAR systems.

2.2 HYPERSPECTRAL SENSOR CONCEPT The hyperspectral sensor combines the following photonic technologies: conventional imaging, spectroscopy, and radiometry—to produce images together with spectral signatures associated with any spatial-resolution element (pixel). The position of spectral imaging relative to related technologies is shown in Figure 2.1.

Hyperspectral Sensor Characteristics

FIGURE 2.1

45

Relationship among radiometric, spectrometric, and imaging techniques.

Data produced by a spectral imager creates a cube, with position along two axes and wavelength along the third, as depicted in Figure 2.2. By proper calibration, the recorded values of the data cube can be converted to radiometric quantities that are related to the scene phenomenology (e.g., radiance, reflectance, emissivity, etc.). The latter provides a link to spatial and spectral analytical models, spectral libraries, and so on, to support various applications. The motion of the spacecraft is used to move the image line across the surface of the Earth, and the CCD for example is read out continuously to provide a complete hyperspectral dataset per line. The imagery for the whole area can be represented in the form of data cubes or a set of images, each image containing information from one wavelength. To illustrate the strength of this technique, the combined spectral/spatial analysis allows detection in an image of optically unresolved (sub-pixel size) objects. Its applications range from Earth remote sensing to early cancer detection. It is obvious why the combination of imaging and spectroscopy is so attractive: conjoining of the two analyses, which can be considered as a data fusion, makes for enhanced image perception. Traditional color film-like imagery touches upon the idea, except that it is based on broad spectral bands and varying spatial resolution, so the details in colors are often achieved at the expense of spatial resolution as exemplified by mixing of red, green, and blue bands into true-color imagery from CCDs.

FIGURE 2.2 Imaging spectroradiometer concept—Data cube composed of individual images recorded over m spectral bands.

46

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Hyperspectral imaging systems have the advantage of providing data at high resolution over a large number of bands, hence are suitable for a wide variety of applications. Their multispectral counterparts, operating over less than 15 bands, are less expensive, yield smaller datasets, and have higher signal-to-noise ratios. They are often tailored to a specific application, so that for other purposes their bands may be suboptimal or even totally unsuitable. However, once the optimal bands have been identified for the task at hand, such a system can provide an overall superior solution. In hyperspectral imaging, the decisive factor is not the number of bands (channels), but the bandwidth—the narrower, the better—and the sequence of acquisition. In other words, a 15-channel system would count as hyperspectral if it covers the band from 500 to 700 nm, with each zone not exceeding 10 nm, while a similar 15-channel system covering the shortwave and visible range and the three IR ranges would count as multispectral. The hyperspectral imaging system breaks down the incoming optical radiation into hundreds of bands and captures specific signatures unique to a scene. Those are processed in a way that permits reliable identification of the objects standing out against their backgrounds, which generate a visual representation of the “hidden” resource. By this means, it is possible to “penetrate” dense foliage and locate shallow underground installations, tunnels, pipelines, and so on. Incidentally, this has led to the term “hyperspectral” being interpreted as hypersensitive to buried or underground objects, which is obviously incorrect. According to the terms of reference approved by the United Nations (UN), monitoring of the environment consists in regular observation of its constituents in space and time, for specific purposes and within the framework of programs worked out in advance. In the context of the vegetation cover, this means observation of its status, control of its current dynamics, forecasts for the future, advance warning of catastrophic upheavals, and recommendations on mitigation or prevention of damage. In the same context, hyperspectral measurements can serve the purposes of understanding: meteorological and climatic factors, pollution of the atmosphere, water bodies and soil, extraordinary human-made and natural situations, and information on land use and reforms. Hyperspectral data have a high commercial potential. Through them, significant savings are expected in the development and operation of monitoring systems, which is a task both complex and expensive, demanding a collaborative effort by the scientific and manufacturing sectors at the national levels. At the same time, it should be noted that the terms of reference of such hyperspectral systems give rise to the following conflicts: • • • •

Between the required coverage and the particularity of the image; Between the required spatial and spectral resolutions; Between the required and real quantization orders of the spectral channel signals; Between the video data volume and the limited traffic capacity of the current satellite communication channels.

Given below are specific features inherent in hyperspectral airborne and spaceborne applications: • Through them, the detection and recognition steps yield accompanying spectral features. These admit lower spatial-resolution levels, which in turn allows recourse to smaller and lighter sensors. The latter can be accommodated on mini-vehicles which, operating in clusters, permit reduction of the observation periodicity for a tract to some hours or less. Another result of the reduced spatial-resolution level is a larger grid unit for the same receiver format. • Hyperspectral data are readily amenable to computerized processing with involvement of geo-informational technologies. This in turn contributes to time-saving and facilitates employment of technician-level personnel. Advanced software has been developed, incorporating geo-informational technologies.

Hyperspectral Sensor Characteristics

2.3

47

HYPERSPECTRAL SENSOR PHYSICS, PRINCIPLES, AND DESIGN

Imaging spectrometers typically use a 2D matrix array (e.g., a CCD), and produce progressive data cubes through successive recording—either of full spatial images, each at a different wavelength, or of narrow image swaths (1-pixel wide, multiple pixels long) with the corresponding spectral signature for each pixel in the swath. Remote imagers are designed to focus and measure the light reflected from contiguous areas on the Earth’s surface. In many digital imagers, sequential measurements of small areas are taken in a consistent geometric pattern as the sensor platform moves, and subsequent processing is required to assemble them into a single image. An optical dispersing element in the spectrometer splits light into narrow, adjoining wavelength bands, with the energy in each band being measured by a separate detector. Using hundreds, or even thousands of detectors, spectrometers can accommodate bands as narrow as 1 nm over a wide wavelength range, typically at least 400–2400 nm (from visible to middle-IR). Hyperspectral sensors cover bands narrower than their multispectral counterparts. Image data from several hundred bands are recorded concurrently, offering much higher spectral resolution than that provided by sensors covering broader bands. Sensors under development, designed to cover thousands of bands with even narrower bandwidth than hyperspectral examples, have a special name—ultraspectral. The most common modes of image acquisition by means of hyperspectral sensors are known as “pushbroom scanning” (electronical) and “whiskbroom scanning” (electro-mechanical). Pushbroom scanners (Figure 2.3) use a line of detectors over a 2D scene. The number of pixels equals that of ground cells for a given swath. The motion of the carrier aircraft or satellite realizes the scan in the along-track direction, thus the inverse of the line frequency equals the pixel dwell time. In a 2D detector, one dimension can represent the swath width (spatial dimension, y) and the other, the spectral range. Pushbroom scanners are lighter, smaller, and less complex than their whiskbroom counterparts because of fewer moving parts. They also have better radiometric and spatial resolution. Their major disadvantage is a large number of detectors amenable to calibration. These imaging spectrometers can be subdivided in Wide Field Imagers (MERIS, ROSIS) and Narrow Field Imagers (HSI, PRISM). These imager-types are conducive to highly repetitive or frequent global coverage when hyperspectral sensors are installed on board LEO satellites. Whiskbroom scanners (Figure 2.4) are on-axis optics or telescopes with scan mirrors sweeping from one edge of the swath to the other. The field of view (FOV) of the scanner can be covered

FIGURE 2.3

Principle of pushbroom scanning.

48

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 2.4

Principle of whiskbroom scanning.

by a single detector or a single-line detector, the carrier’s motion implementing the sweep. This means that the dwell time for each ground cell must be very short at a given instantaneous field of view (IFOV), because each scan line consists of multiple ground cells to be covered. Whiskbroom scanners tend to be large and complex. The moving mirrors bring in spatial distortions to be corrected by preprocessing before delivering the information to the user. An advantage of the whiskbroom scanners is that they have fewer detectors subject to calibration, as compared to other types of sensors. Well-known examples are AVHRR, Landsat, and SeaWiFS. Two approaches to hyperspectral sensors were launched and operational for many years in the new millennium: the dispersion approach (e.g., EO-1) and the Fourier Transform approach (e.g., MightySAT). • Dispersion. By this technique, the spectral images are collected by means of a grating or a prism. The incoming electromagnetic radiation is separated under different angles. The spectrum of each ground pixel is dispersed and focused at different locations of the 1D detector array. This technique is used for both image acquisition modes (pushbroom and whiskbroom scanners). Hyperspectral imagers mainly use gratings as the dispersive element (HSI, SPIM). • Fourier Transform Spectrometers (FTS). Spatial Domain Fourier Transform Spectrometers like the well-known sensors SMIFTS or FTHSI on the MightySat II spacecraft use the principle of the monolithic Sagnac interferometer. Unlike conventional FTS, the Earth Observation spectrometers in LEO operate with fixed mirrors. The optical scheme distributes the interferogram (spectrum) along one dimension of the detector, the other dimension representing the swath width.

Hyperspectral Sensor Characteristics

49

Classical hyperspectral imaging is performed by one of two basic principles: scanning by a pushbroom approach or by a tuneable filter approach, for example using an Acousto-Optic Tuneable Filter (AOTF). In a pushbroom scanner, a line instrument is used where the light is diffracted into different color/wavelength components by means of a spectrograph onto a focal plane array. Unlike it, in the tuneable filter approach, instead of scanning the image stepwise in one spatial dimension, the entire spatial images are taken, as the camera scans through the different wavelengths. This is achieved by placing the AOTF between the image and the camera [9]. AOTF is solid-state device comprising a crystal filter whose transmission wavelength is controlled by attached piezoelectric transducers. The transducers create pressure waves in the crystal which changes its refractive index. These transducers are controlled by careful manipulation of the frequency of an applied electric field to vary the transmission wavelength of the filter in a stepwise manner. An image captured at each wavelength step is read into memory to form a slab of the hypercube. The data are then processed in order to produce the desired output image. The spectrograph in a pushbroom system provides better spectral resolution than the tuneable filter system, but at a cost of speed, while the latter device, with no moving parts, builds up the hypercube more swiftly. For detailed studies of the Earth’s surface in the reflection spectral band of 400–2400 nm, the onboard hyperspectral equipment should simultaneously provide high spatial resolution of 1–5 m, high spectral resolution of 5–10 nm in tens and even hundreds of channels, a swath of up to 100 km, and a signal-to-noise ratio ≥100. For devices to approach these requirements approximations have been made. One remedy consists in replacing the fixed-parameter sensors by adaptive systems. A major consideration in developing the latter systems is that they would operate under widely-varying conditions, including parameters of the atmosphere, characteristics and types of the Earth’s surface, the space platform motion, and attitude. These variations are stochastic, which tends to impair both the efficacy of the observation process and the quality of the obtained images. Under the adaptive regimes, the information parameters of the system are selected and adjusted in real time for the problem in question. Among the adjustable information parameters are the spatial and spectral resolutions as well as the signal quantization width. The performance level of the observation missions can be raised to a qualitatively new stage by implementing the advanced operation modes for the onboard equipment. One more prerequisite to this end is to reduce the volume of the transmitted imagery, combined with prevention of information losses. Redundancy should be curtailed by a three- or even fourfigure factor, which maybe unfeasible by the traditional means alone. The proposed solution consists in real-time quality assessment of the observation data and selection of the informative onboard channels, to screen out the data with low informational content. This approach not only enables redundancy reduction, but also enhances the information output. At present, the spectral intervals, their number, and their layout are chosen at the design stage and remain practically unchanged afterward. New methods of interval selection, currently under development, are based on analysis of the spectral reflection characteristics. Unlike the conventional procedure, the intervals and their limits are selected through a compromise between the required spectral and spatial resolutions. This is implemented, for example, in Full Spectral Imaging (FSI)—a method for acquiring, preprocessing, transmitting, and extracting information from full-spectrum remote sensing data by the “spectral curve” approach, instead of the current “bytes-per-band” one [10]. By virtue of the bandwidths’ optimization, FSI admits simplified instrument characteristics and calibration as well as reduced data transmission and storage requirements. These improvements may be accomplished without loss of remote sensing information. FSI is neither “hyperspectral,” nor “superspectral” or “ultraspectral” imaging; it is an end-to-end system which involves the whole technological sequence, from the observation technique to the data processing. FSI belongs to the class of hyperspectral-type imagers based on a dispersive element and a pushbroom image acquisition system. Instrument throughput, spatial coverage, and spectral response can be enhanced using multiple focal planes. The multiple entrance slits would be optically connected to several spectrometer-and-detector combinations. Each spectrometer would have identical optical elements except for the dispersion characteristics of the grating and the detector.

50

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

An important task is the screening of received hyperspectral data to obtain a true representation of the Earth’s surface. The information involves random, systematic, as well as system-defined distortions caused by the atmosphere, curvature of the Earth, displacement of the hyperspectral equipment relative the Earth’s surface at the moments of imaging, and physical characteristics of the sensors and channels. To remove these multiple distortions, a number of corrections should be applied, based on radiation, radiometric, geometrical, and calibration techniques. Radiation correction is related to adjustment of the EM energy amount as received by each sensor, thereby compensating for diverse atmosphere transparency to the different frequency bands. Radiometric correction represents elimination of the imaging system-defined distortions inserted both by the sensors and the transceivers. Geometrical correction, or image transformation, is intended to remove the distortions due to the Earth’s curvature and rotation, inclination of the satellite’s orbit with respect to the equatorial plane and, for high-resolution images, the terrain relief. This correction can be done automatically, with the hyperspectral sensors position known. More precise transformation and image linking to a specific coordinate frame is usually achieved by interactively setting the reference points. In the transformation process, the pixel coordinates conversion to a new raster frame may cause some changes in objects shape. In editing and joint processing of the different imaging data types, as well as the images of the same terrain taken at different times, it is internationally accepted to use, as an exchange standard, a so-called orthoplan projection. Calibration consists in converting the dimensionless data, as obtained by the sensors of various spectral bands, into true normalized values of the reflected or emitted energy.

2.4 HYPERSPECTRAL SENSOR OPERATIONAL MODES The recent decades have been marked by the fast advance of the hyperspectral techniques. Development of the technical means for hyperspectral imaging went all the way from simple lab devices with limited capabilities to sophisticated space complexes, which enable excellent imaging of the Earth’s surface simultaneously in hundreds of the narrow spectral bands, and successful downloading of this information to the Earth. Today, there are three basic methods of acquiring the hyperspectral images of the Earth: ground-based, airborne- and spaceborne imaging. Each method has peculiarities, advantages, shortcomings, and a range of problems of their own. Imaging method depends on the sensors’ structural features imposed by their operating conditions—on the ground, in the atmosphere, or in space. The maximum efficiency can be attained when all the tools are available for working in all three spheres, as well as the opportunity of choosing the optimum way to achieve the goal. None of the cited methods duck challenges offered by the path of progress, and their simultaneous improvement brought an impressive rise in hyperspectral techniques. Some general descriptions and results obtained by the hyperspectral techniques in the different hand-held, truck-mounted, airborne, and spaceborne applications are presented here by way of illustrating some of the points made, but not a substitution for the wealth of details provided by attending chapters (Figure 2.5).

2.4.1 ground-based hyPersPectraL imaging Hyperspectral imagery is generated by the instruments called imaging spectroradiometers. The development of these sophisticated sensors involves the convergence and synergy of two related but distinct technologies: spectroscopy and remote imaging of the Earth. Spectroscopy is a study of emission, absorption, and reflection of light by the matter as a function of radiation wavelength. As applied to the field of optical remote sensing, spectroscopy deals with the spectrum of sunlight that is diffusely reflected (scattered) by the different materials on the Earth’s surface. Ground-based or lab measurements of the light reflected from the test material became the first successful application of the spectrometers or spectroradiometers.

Hyperspectral Sensor Characteristics

FIGURE 2.5

51

Landscape fantastic HS control.

On-ground hyperspectral imagers are ideal tools for the identification of obscured or resolutionlimited targets, defeating camouflage, and recognition of both solid and gaseous chemicals. There are manufacturing processes that can make use of MS or hyperspectral data, which include inspection of color and paint quality, detection of rust, or the flaw detection in thin film coatings, and so on. Hyperspectral measurements made it possible to derive a continuous spectrum for each pixel. Once the adjustments for the sensor, atmospheric, and terrain effects are applied, these image spectra can be compared with field or lab reflectance spectra, in order to recognize and map surface materials, for example, specific types of vegetation. Collections of spectral curves for both the materials and the objects, routinely measured under lab or field conditions, are stored in the spectral libraries. These curves can be used for classification or automatic identification of objects and materials. Until recently, the imagers’ resolution capability was restricted due to the limitations, imposed both by the detector design and the requirements to the data storage, transmission, and processing. Recent advances in these areas have contributed to the imagers’ design, so their spectral ranges and resolution are now comparable with those of the ground-based spectrometers. Surface Optics Corporation (SOC), which is specializing in the characterization and exploitation of the optical properties of surfaces, has developed a wide range of commercial products including hyperspectral video imagers and processors. Its SOC-700 family of hyperspectral imaging systems is devised to address the real-time processing requirements of various applications, such as machine vision, biological screening, or target detection. The hyperspectral imager (Figure 2.6) is a highquality, portable, and easy-to-use spectral imaging instrument, radiometrically calibrated, with a software intended for both the analysis and viewing. The imager captures a 640 × 640-pixel image that has 120 spectral bands deep in as fast as four seconds. Hyperspectral sensors can acquire and process up to 100 lines of 12-bit data for all 120 bands and return an answer, to be used in controlling the process, screening the cell samples, or telling apart friend from foe. The hyperspectral analysis package provides the tools for serious spectral work, including complete radiometric calibration, reflectance calculations, matched filtering, and atmospheric correction. The sensor is set up within minutes and can withstand rugged outdoor environments. Since its inception, it has been used under rainforest and desert conditions and has been drop-tested from a 4-ft height. Design and tests of hand-held hyperspectral devices has improved real-time detection and quantification for a wide spectrum of applications. An impressive hyperspectral imager, produced

52

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 2.6 HS target detection system on a field trial.

by Bodkin Design, can capture both the spectral and the spatial information of the scene instantly, without scanning [11]. It is designed for high-speed hyperspectral imaging of transient events and observation from moving platforms. The device has no moving components, making the system immune to mechanical failure. Its unique design simplifies the detector read-out and the optical system, and provides a fully registered, hyperspectral data cube on every video frame. It is ideally suited for airborne, hand-held, and vehicle-mounted systems. The number of hyperspectral sensor applications in various areas of science and technology constantly improves. Thus, Terrax Inc. yields the Theia—a low-power, lightweight hyperspectral imaging system with automated image analysis and data optimization for real-time target detection and hyperspectral data comparison. It consists of cameras and a hyperspectral real-time imagecapture and decoding subsystem. The Theia is designed to fit within the size, weight, and power envelope of a portable environment as well as fixed locations. It is robust and includes several on-board real-time target detection and data optimization algorithms based on a calibration library of 100 substance cubes. The system called foveal hyperspectral imaging [12] employs a hyperspectral fovea with a panchromatic periphery. This approach results in a low-cost and compact hand-held hyperspectral imager suitable for applications such as surveillance and biological imaging. In these applications, the panchromatic peripheral image may be used for situational awareness and screening of the potential threats of the hyperspectral fovea to that region of interest for recognition or detailed characterization. As was demonstrated in the applications, the real-time capability of the foveal hyperspectral imaging shows particular promises including surveillance, retinal imaging, and medicine. The hyperspectral imager is a compact asymmetric anamorphic imaging spectrometer covering broad spectral regions available from a specialty manufacturer. The complete imaging system is contained in a variety of packages suitable for deployment in a mobile vehicle environment or with the imaging head remote from the processing stack, as in building’s perimeter security systems. The detection capability and corresponding display varies with the model and the database of cubes representing calibration for specific chemical or material signatures. Detection and determination can reliably work over distances from a few feet to 300 ft. The Norwegians have built an interesting technology demonstrator for hyperspectral target detection [13]. The ground-to-ground demonstrator system was also developed, which is currently being used in tests. The package contains a HySpex VNIR hyperspectral camera, a line scanner monochrome camera, and a motorized turntable mounted on a tripod. The rotation axis and the linear FOV are vertical, so that the cameras scan around the horizon. Shown in Figure 2.6, the

Hyperspectral Sensor Characteristics

53

hyperspectral system was success tested on a field trial in northern Norway. The system is designed for detection of the small targets, as well as for search and rescue operations. Concerning the last option, it should be noted that in many cases there is no need to identify a specific target—what is needed is to detect the presence of a target with unknown a priori characteristics. Search and rescue operations provide good examples of such an approach. For instance, consider a hiker lost in the mountains or a boat on the high seas to be promptly detected. Rather than to conduct an extensive analysis of hyperspectral data cubes, which is quite a time-consuming task, a cursory technique based on “anomaly detection” can be applied. Some researchers working with ground remote sensing systems have been carried out using sensors attached to long hydraulic booms hoisted above the crop canopy from the ground (the University of Illinois). The images collected from such a close distance have resolutions much higher than those taken from aircraft or satellites. Other ground-based systems use vehicle-mounted sensors that control variable-rate applicators in real time. For example, the remote sensors distinguishing the weeds from the crop are mounted on sprayers that change the application rate of herbicides, to be applied on the go. This form of remote sensing technology, called machine vision, is now in wide use. Most of the hyperspectral imagers operate in the visible to shortwave IR bands. Hyperspectral imaging in the IR spectrum range has multiple applications, and it is an urgent problem to develop the relevant instrumentation. Many industrial firms develop and turn such a production out. One similar product that can operate both on the ground, and onboard a plane, is presented below. The Aerospace Corporation has designed and built the state-of-the-art narrowband hyperspectral Fourier Transform IR (FTIR) scanner—Spatially Enhanced Broadband Array Spectrograph System (SEBASS [49]), which offers some unique capabilities within the hyperspectral remote sensing arena. SEBASS can collect data either from the transportable ground-based sensor, or from a low-flying aircraft, depending on the application at hand. The SEBASS capability in the mid-wave and longwave IR is intended to remotely identify materials in the 3 to 13 µm “chemical fingerprint” spectral region. The hyperspectral sensor also provides high-resolution temperature data. SEBASS captures 128 spectral bands in the IR spectral range with its IFOV of 1 mrad, and FOV of 128 mrad (7.3°). Its typical remote IR products include spectrally and radiometrically calibrated, atmospherically corrected hyperspectral data cubes. SEBASS has three operational modes [49]: • Vehicle Mounted Measurements—four mobile 2D scanning longwave IR FTIR vehicle mounted sensors. The vehicles can be driven to the site of interest to collect close-range hyperspectral data from 8 to 12 microns. The scanning FTIR sensors have been successfully deployed on over 100 ground-based data collections. • Hand-held Measurements—provide in situ ground spectral measurements using hand-held instruments to verify material identifications derived from airborne measurements. • Laboratory Measurements—to assist in material characterization of samples taken from regions of interest, establish ground truth, and improve interpretation of remote sensing data. Incidentally, as was noticed by many researchers, the best results are those obtained by sharing the data as obtained by various hyperspectral sensors at different positions with respect to the target [14]. In this way, for example, sugar beet disease was detected by combining multitemporal hyperspectral remote sensing data, as provided by the different airborne, tractor and hand-held spectroradiometers. Imaging spectrometry data are well established for detailed mineral mapping from airborne and spaceborne systems. Overhead data, however, have substantial additional potential when used together with ground-based hyperspectral measurements. United States researchers used the hyperspectral scanner system to acquire airborne data, outcrop scans, and to image boxed drill core and rock chips at approximately 6 nm nominal spectral resolution in 360 channels from 400 to 2450 nm. Analysis results using standardized hyperspectral methodologies demonstrate rapid

54

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

extraction of representative mineral spectra and mapping of mineral distributions and abundances. A case history highlights the capabilities of these integrated datasets for developing improved understanding of relations between geology, alteration, and spectral signatures in three dimensions.

2.4.2 airborne hyPersPectraL sensors Up until 2000, a fleet of over 40 airborne sensor systems gave the mature form of the future of data acquisition opportunities. During most of the 1990s, hyperspectral imaging had been an area of active research and development, so the images were available only to researchers. With the recent appearance of commercial airborne systems and with applications in resource management, agriculture, mineral exploration, and environmental monitoring, hyperspectral imagery is poised to enter the mainstream of remote sensing [15–17]. But its effective use requires an understanding of the nature and limitations of both this data type and various strategies involved in its processing and interpretation. Like the ground equipment, the airborne hyperspectral scanners are also designed to measure the radiance scattered by the Earth’s surface, in the given spectral band—any pixel of the acquired image simultaneously contains both spatial and spectral information on the terrain. The input radiation flux is split into components, according to a wavelength. For each waveband, certain lines of the matrix are reserved. These data are digitized and written to storage of the unified control, data storage, and power supply module. For spatial linking of the collected imagery, the inertial navigation system (INS) is involved in the post-flight processing of its own data together with those of the imaging scanner, to attribute geographic coordinates to each pixel. More advanced survey complexes are supplied with navigation systems which have the relevant software to control the airborne imagery process. Thus, along with imagery planning and recording, the system also controls both the aircraft and the imaging sensors during the flight. The estimation of the navigation and imagery precision is carried out in real time. To the credit of such imagery control systems and their high quality and reliability is the fact that about 150 of them are operational throughout the world today. Companies-producers of such equipment in many countries have mastered manufacturing of the guide-beams for the hyperspectral scanners, including those to operate in visual and near-IR (VNIR), shortwave IR (SWIR), medium-wave IR (MWIR), and thermal IR (TIR) bands. The airborne scanners, which are currently becoming operational, are notable for their perfection and variety, the number of their different types supposedly exceeding one hundred. The scanning hyperspectral sensors of the leading companies have reliable, stable, and functional components; a software handy for operation; high capacity of digitizing and readout; high-precision and efficient optics; stable and reproducible parameters of the sensor calibration, and so on. The customers are often supplied by the companies not just with the imaging hardware, but with the complex imaging technologies. Such integrated digital airborne systems represent the off-the-shelf solutions for obtaining the final product. Generally, each hyperspectral scanner consists of a sensor (optico-electronic unit); power supply; data accumulation and control unit; vibration damping platform; operator display; GPS receiver; inertial measurement unit; and data processing device. A matter of priority is also to maintain functional completeness of the hyperspectral scanners’ spectrum, that is, a possibility to use them in all the applications somehow connected with the spectral analysis. It is only in this case that the quantitative assessments can be made of the spectral characteristics, pertaining both to the underlying surface and specific ground objects, which, in turn, are extremely important for obtaining trustworthy results at the stage of applied analysis. This is why the companies are heavily investing in the development of their own optico-electronic tract, including CCDreceivers with high signal-to-noise ratio, stable radiometric calibration, and so on. The equipment design and the operational software make it mostly possible to make measurements at the moment of imaging and to account for it later on.

Hyperspectral Sensor Characteristics

55

Airborne hyperspectral imaging ensures object recognition is made by their physicochemical makeup and enables one to identify plant species and vegetation status; while keeping under control the natural habitat of the different plants including weeds and unique plants; as well as spot vegetation and soil disturbances and find wetlands and salt disturbed soils. Aircraft, drone, helicopter, and airships—each of them has some useful advantages as a platform for remote sensing systems. Aircraft can fly at relatively low altitudes, thus admitting a sub-meter sensor spatial resolution. The airborne sensors can be used more flexibly because of the variation of height (flight level) and the forward scan velocity. Therefore, spatial resolution and swath width can be easily adapted to the task requirements in the course of flight. Aircraft can easily change their imagery schedule to avoid weather problems such as clouds, which may block a passive sensor’s FOV, and to allow for the sun illumination. Sensor maintenance, repair, and configuration changes are easily applicable to aircraft platforms. There are no bounds for an aircraft but the state borders. At the same time, for a low-flying aircraft with a narrow sensor’s FOV, a number of flybys are needed to cover a large ground area. The turnaround time it takes to get the data to the user is delayed, due to necessity of landing the aircraft before transferring the raw imagery to the data processing facility. Besides, the hyperspectral data can be easily and efficiently integrated with those obtained by the airborne LiDAR system, or the sensors operating in other spectral bands. The hyperspectral scanners are the most advanced and sophisticated airborne optical devices of the Earth remote sensing. Here we name but a few of the state-of-the art hyperspectral airborne imagers with excellent characteristics, such as AVIRIS, HYDICE, AISA, HyMAP, ARES, CASI 1500, and AisaEAGLET and give basic specifications of the most interesting among them. The Avionic Multisensors hyperspectral System, manufactured by the SELEX Galileo company, includes four hyperspectral/MS optical heads (e.g., VNIR, SWIR, MWIR, and TIR) which provide, in different configurations, a spectral coverage from the visible (400 nm) to the TIR (12,000 nm) band; the inertial platform with integrated GPS (INS/GPS) to take records of the aircraft position and attitude; the Instrument Control Unit/Pre-Processing Computer to control the optical head and store data in internal HD memory. The “modular” approach allows, with just a change of mechanical interface, a flexible arrangement of instrument accommodation, making it suitable for use on the different platforms, including unmanned aerial vehicles (UAVs) and ultra-light aircrafts. The hyperspectral system can be applied both in airborne and ground operations, although its basic configuration is designed for airborne platforms, with pushbroom scanning to build up the spectral data cubes. To derive geo-registered images, the flight data coming from a dedicated GPS/ INS unit are continuously logged in sync with the hyperspectral data. In the ground system, the same optical head is used as in the airborne system, as well as the scanning platform synchronized with the image acquisition. In this mode, the instruments can be used as “static” cameras for applications, where the linear platform movement needed for pushbroom mode is not available. The development follows a modularity principle aimed at creation of a flexible system to be utilized on manned or unmanned platforms for different applications, such as UAV, light aircraft, medium-range aircraft for maritime and coastal surveillance and patrol, and multi-mission maritime aircraft for high-altitude surveillance. SPECIM airborne hyperspectral sensors provide market leading solutions for remote sensing, from small UAV systems to full-featured commercial, research, and defense tools. SPECIM brought to the market the AisaEAGLET hyperspectral sensor, particularly designed to meet space and weight limitations of UAVs and small piloted aircraft (Aisa stands for Airborne Imaging Spectrometer for Applications). This hyperspectral sensor is the most compact and complete airborne imaging system to acquire full, contiguous VNIR data with a high spatial resolution of 1600 pixels and excellent sensitivity even under low-light conditions, with the imager’s spectral range of 400–1000 nm and spectral resolution of 3.3 nm (up to 200 spectral bands). Its FOV is ∼30°, IFOV—0.02°; resolution on the Earth’s surface ∼0.35 m from 1000-m altitude. The system has a total mass of 10 kg, including the hyperspectral head (3.5 kg), compact data acquisition computer, and GPS/INS unit. The sensor

56

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 2.7 Optical head of AisaEAGLET hyperspectral sensors destined for operation on small aircrafts.

head, shown in Figure 2.7, contains a wavelength and radiometric calibration file. System output—12 bits digital, SNR in range of 130:1–300:1 (depending on the band configuration), power consumption 500:1) and image quality setting the industry standard. Lab calibration and operational system monitoring ensures the imagery to be calibrated as required for demanding spectral mapping tasks. Geo-location and image geocoding are

Hyperspectral Sensor Characteristics

57

based on differential GPS and INS. Hyperspectral sensors IFOV—2.5 mrad along track, 2.0 mrad cross track; FOV—61.3° (512 pixels); ground IFOV is 3–10 m for typical operational range. The new AISA hyperspectral Imager in dual mode is capable of acquiring 492 bands of 2–6 nm bandwidths across the spectral range of 395–2503 nm, at pixel sizes between 75 cm and 4 m. It is mounted on a twin-engine Navajo along with a LiDAR system with 10-cm vertical accuracy and a digital camera. A known HYDICE sensor collects data of 210 bands over the range of 400–2500 nm with FOV of 320 pixels wide at the IFOV, projection of which on the ground (pixel size) of 1–4 m, depending on the aircraft altitude and ground speed. Let’s give some more examples of the current-day hyperspectral sensors which contributed in raising the technological level in the area. Successfully tested in the United States was the hyperspectral sensors onboard the reconnaissance UAV designed for spotting the tanks, missile launchers, and other camouflaged defense equipment located on the colorful background. The system developed by the TRW Company provides imaging in 384 working spectral bands. Another example is the Earth Remote Sensing (ERS) lab, with high spatial and spectral resolution, operated worldwide and lately, as an example, successfully commissioned in Kazakhstan. Its equipment consists of a widescreen 136-Mpixel UltraCAM X camera—one of the most powerful airborne devices of its kind in the world market—and a high-resolution CASI 1500 hyperspectral scanner, with 288 programmable channels in the visible and IR bands. By its technological level, the equipment meets standards of the satellite-borne ERS, the only difference being the carrier involved. The advantage of placing the equipment onboard the aircraft lies in the high resolution on the terrain (3–50 cm) and in highly accurate 5 cm linking, both in plane and in height, as ensured by the onboard navigation system data. The coverage resolution achieved by the airborne hyperspectral imaging is within 25–400 cm. At the maximum flight altitude, an 8 km swath can be covered, as with the high-resolution OrbView-3 satellite. To cope with huge data streams, the onboard equipment also includes the 13-processor supercomputer with 4 Tb memory.

2.4.3 sPaceborne imaging Big hopes are pinned on the imaging from space. For a long enough time, the MS systems performed onboard the spacecraft, generating several images in the separate wide spectral bands, from visual to IR. Until now, MS imaging from the new generation spacecraft (GeoEye: 1–4 spectral zones, RapidEye: 5 zones, and WorldView: 2–8 zones) was of a great practical interest. Nowadays, there are opportunities of developing spaceborne hyperspectral equipment for observations in the optical range, with high informational and operational characteristics, which would enhance a solution level of the space monitoring tasks [18–22]. Satellite platforms enable placement of various sensors for observation and provide coverage of the Earth’s surface and periodic monitoring of the areas of interest during the satellite revisits. The image resolution is defined by the satellite orbital parameters, and its altitude in the first place. For satellites, there are no political borders, so they can sweep any part of the globe, regardless of the attitude of the states the onboard imaging equipment flies over. The basic weakness of the spaceborne imagery is its multi-hundred million dollars costs to develop the system as well as ground support facilities, while keeping in mind the spacecraft’s relatively short operating life of about 5 years, or less. Decrease of the satellite mass can result in substantial cost saving, primarily due to less powerful and low-cost launchers to be involved in putting the satellite into orbit. This is why the concept of small satellites-based hyperspectral imaging was developed and implemented by means of mini-satellites (100–500 kg). Basic characteristics of the space-based hyperspectral instruments are presented in Table 2.1. The list includes both flight-proven devices and those under development for future operations in space [23–35]. Most of them are devised for LEO small satellite missions, although placing possible hyperspectral sensors onboard the large satellites and satellites on geostationary orbit is also considered [36]. The first fourteen rows in Table 2.1 pertain to hyperspectral satellite with a swath on the Earth surface of 5–20 km, number of spectral bands ∼200, and spatial and spectral resolution

58

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

TABLE 2.1 Space-Based Hyperspectral Mission Overview: Operational and Planned Instruments Instrument (Satellite)

Altitude (km)

Pixel Size (m)

HSI (SIMSA) FTHSI (MightySatII) Hyperion (EO-1) CHRIS (PROBA) COIS (NEMO) ARIES-1 (ARIES-1)

523 565 705 580 605 500

25 30 30 25 30 30

UKON-B Warfighter-1 (OrbView-4) EnMAP

400 470

20 8

675

30

Number Bands

Spectral Range (nm)

220 256 220 19 210 32 32 32 256 200 80 92 108

430–2400 450–1050 400–2500 400–1050 400–2500 400–1100 2000–2500 1000–2000 400–800 450–2500 3000–5000 420–1030 950–2450 400–2500

Spectral Resolution (nm) 20 10–50 10 1.25–11.0 10 22 16 31 4–8 11 5–10 10–20 10

IFOV (µrad)

Swath (km)

47.8 50 42.5 43.1 49.5 60

7.7 13 7.5 17.5 30 15

50 20

15 5

30

30

HypSEO (MITA)

620

20

∼210

40

20

MSMI (SUNSAT)

660

15

400–2350

10

22

15

PRISMA

695

30

∼200 250

400–2500

30

425

4

400

400–2500

200 8

380–2500

10

80

∼10 145

430–910

20

30

120

HyspIRI SUPERSPEC (MYRIADE) VENµS Global Imager (ADEOS-2) WFIS (like MODIS)

∼700 720

60 20

720

5.3

12

415–910

16–40

8

27.5

802

250–1000

36

380–1195

10–1000

310–1250

1600

705

1400

630

400–1000

1–5

2000

2400

of 10–30 m and 10–20 nm, respectively. The SUPERSPEC and Vegetation and Environment New Micro Satellite (VENµS) sensors represent multispectral imagers of the next generation; it was put in the table for comparison with previous hyperspectral instruments. The focal plane of multispectral cameras consists of a multi-line array, one for each spectral band with relative narrowband spectral filters. Along with the narrow FOV hyperspectral instruments, operational interests would also include some specimens of the wide FOV hyperspectral imaging systems. Examples of such hyperspectral missions are shown in the last two rows of Table 2.1. It is obvious that space applications of hyperspectral remote sensors are under constant development. An indicative example of remote sensor advancement is the above mentioned joint Israeli/French scientific and technological project VENµS, launched in 2017. It is designed for high-definition photography of agricultural tracts as a means of ecological control. VENµS launch weight was 265 kg, and its expected lifetime in space was more than 4 years. It has superior parameters in terms of the number and width of its bands, spatial resolution, imaged strip width, mass-energy, and information capacities. The 90-W camera weighs 45 kg and discerns the objects down to 5.3 m in size from 720-km altitude, scanning across 27.5-km swaths. Photographing will be carried out in 12 narrow spectral channels in the 415–910 nm range, with each band varying from 16 to 40 nm. In this mission, most satellite and sensor design parameters are optimized.

Hyperspectral Sensor Characteristics

59

Let us briefly review the worldwide state-of-the-art in the field. There are a small number of experimental operational spacecraft equipped with hyperspectral sensor, viz.: EO-1 (under the NASA program) and MightySat II.1, aka Sindri (under the US Air Force program); two other defense projects—ORBView-4/Warefighter and NEMO—are at various stages of readiness. The information capacity for different spectral bands was assessed by processing the measurement data as obtained in simultaneous imaging of the test objects by hyperspectral sensors and traditional optical technique (EO-1 with Landsat 7 and Terra; MightySat with the Keyhole visual reconnaissance satellite). Investing in commercial satellites can be a risky business. TRW’s Lewis satellite with hyperspectral sensors was lost shortly after launch in 1997. In the same year, EarthWatch lost its EarlyBird satellite 4 days after launch. In spite of the risk, every firm feels obligated to incorporate such systems in future spacecraft meant for the Earth remote sensing. In Russia, for example, there is a metrological basis for the subject-related processing of the spectrometer data. Specific prototypes of the hyperspectral equipment were developed by the AFAR scientific production enterprise and the Reagent scientific-technical center [37]. The operation of the Yukon-B video-spectrometric system, with mass of 1.5 kg, is based on the optoacoustic effect in the visible band (500–800 nm). The overall number of zones is 256, from which 10 to 100, with width of 4–8 nm, are program-selected. Spatial resolution at 400-km altitude will be 20 m, and the frame size of the surface area—15 × 12 km, with one zonal image size amounting to 4–5 Mbit. Yukon-B is a component of the projected Yukon-UVIT complex designed to cover the UV, visible, and thermal spectral bands. As a platform for Yukon-B, the service module of the ISS Russian segment can be used, as well as small D33 spacecraft of the Monitor-E model manufactured by the Khrunichev Plant, and Kondor-E of the Scientific Production Enterprise of the Machine Building. Another Russian project is the Astrogon mini-satellite, under development at the Research Institute of the Electric Machinery at Istra, near Moscow, as a part of the Gazprom Trust aerospace monitoring program. The Reagent’s hyperspectral video-camera has the 700-channel capacity (with margin for increase up to 1000), with a spatial resolution of 5 m and adaptive control along both the spectral and spatial coordinates. The programmed control system permits the selection of the resolution level from 0.5 to 100 nm in the spectral bands of 200–900 nm and 1000–1400 nm, as well as one of two possible signal polarization modes. For the signal space-frequency transformation, both rearrangeable optoacoustic filters and micro-channel photon detectors are used. Total mass of the imaging equipment is 4 kg, with power consumption of 15 W. Designed precision of the geometric matching of the images is 5 m. The camera control system makes it possible to carry out both instrumental and virtual stereo-imaging. Two more developments, pertaining to the hyperspectral field, are at different stages of implementation in the Russian space industry, viz.: • On the basis of hyperspectral technology of the mentioned Astrogon project, tested in the experiments onboard planes and helicopters, under consideration is a joint development with Germany of the mini-satellite formation for space monitoring of the wildfires, volcano eruptions, and other calamities. Involved in the project is the IR technology, applied by the German party on the small BIRD satellite. The monitoring would make it possible to discern and locate the sources of the excess energy flux of both natural and anthropogenic origins. • The Vavilov State Optical Institute (St. Petersburg, Russia) has developed a compact onboard hyperspectral camera for the small spacecraft, to receive simultaneously in ∼100 spectral channels of the visual and near-IR bands (200–1000 nm) a 2D distribution of the brightness field with angular resolution ∼20″, which corresponds to 100 m spatial resolution on the Earth’s surface for 1000-km orbit. The instrument provides detailed spectral information on the small gaseous constituents of the atmosphere, status of the inland waters, and vegetation and soil specifics.

60

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 2.8

Artist’s impression of TacSat-3 satellite in orbit.

A remarkable achievement in the hyperspectral-assisted space exploration was the successful launch in May 2009 of the multipurpose TacSat-3 US Air Force satellite from the NASA Wallops Flight Facility on the Virginia coast [38]. The satellite, shown in Figure 2.8, orbiting at a 425 km altitude, is required to deliver hyperspectral images within 10 minutes after an image acquisition. The main component of its specialized equipment, the ARTEMIS (Advanced Responsive TacticallyEffective Military Imaging Spectrometer), designed to operate in 400 spectral intervals in the 400– 2500 nm band (the entire visible and very-near IR), is an advanced version of the Hyperion scanning spectrometer (NASA, 2000), which performs in 220 intervals with spatial resolution of 30 m/ pixel. Space photos with such a resolution form the basis for scanning coverage of the popular geointerfaces, like Google Earth. Besides the spectrometer, the ARTEMIS system includes a telescope and a signal processor with 16-Gb onboard memory. The secondary mirror of the telescope is integrated into the focal plane light-receiving matrix, common for all the bands, unlike its separate counterparts in the prior generation of hyperspectral detectors. This improvement enabled a simpler design of the telescope optical setup, easier data processing, and a cost saving. The duration of the TacSat-3 satellite demonstration mission was just under 3 years. Let’s return to the hyperspectral missions and look at some problems connected with the development of the hyperspectral satellites and their hyperspectral sensors. Energy resolution, defined as a number of resolved levels of the object’s brightness, or of its image illumination, is chosen in accordance with required signal-to-noise ratio. It is well known that detection and recognition probability rise together with the ratio increase. Besides, consideration should be taken regarding the interconnection between spatial, spectral, and energy resolutions, which takes place in live hyperspectral sensors. For example, if higher spatial resolution is achieved by reducing the size of the image element, then the element would get a lesser share of the energy, needed for splitting by operating spectral ranges and obtaining the required signal-to-noise ratio in each of them. This is why various combinations of the basic characteristics can be found in the real devices, depending on the task at hand. The true parameters—bandwidths and instantaneous fields of view (IFOV)—of existing hyperspectral imagery systems in the visible spectral range are presented in Figure 2.9. The interrelation between spectral and spatial resolutions, as calculated for a hyperspectral instrument

Hyperspectral Sensor Characteristics

FIGURE 2.9

61

Bandwidth/IFOV relation for visible range.

with a 20-cm aperture, suitable for operation on the small satellites, is drawn by a solid line. As seen, there is a good conformity between the theory and practice. The major obstacle in developing hyperspectral sensors is a well-known difficulty with designing the relatively cheap high-sensitive photo-detectors, operating in a wide spectral range, with highspatial, spectral, and time resolution. This is also relevant with regard to constructing the efficient, durable, compact cooling systems for the radiation receivers, as well as to some more scientific and technical problems. Not all the attempts made thus far to design hyperspectral sensors as a Fourierspectrometer were successful, since the requirement of the real-time performance failed to be met. Another field of application for the small satellites may be the installation of a compact coarseresolution imager to correct imagery, obtained by other sensors, for atmospheric variability caused mostly by water vapor and aerosols. Such an Atmospheric Corrector (LEISA AC) was used in the EO-1 land imaging mission. Using wedged filter technology, the AC provided the spectral coverage of 890–1580 nm with moderate resolutions (spectral ∼5 nm and spatial ∼250 m at nadir). Conceiving the onboard communication system with a high-rate download of as many simultaneous images as there are channels, powered by additional solar panels, it should be taken into consideration that enhancing spatial resolution would imply the weight increase. Likewise, adding more channels would also result in more weight as well as in more power consumption. Moreover, with the orbit altitude increase (e.g., from 400 to 1200 km), or with wavelength increase (from visible to IR) the imager aperture diameter should also be enlarged. For higher ground resolution, the satellite dimensions are defined not only by dimensions of the communication and power units, but also by those of the imager. Thereby, the requirements to the transmitting antenna, optical system, total power consumption, and so on, would impose operational constraints on the mission design, which should be eventually translated into restrictions on the spacecraft dimensions and mass. Given the existing physical and technological constraints, the resulting ambiguity as to the satellite dimensions urged us to determine the estimation criteria for the satellite’s mass versus spatial resolution and channel array. Depicted in Figure 2.10 are feasible combinations of the hyperspectral imagers with LEO observation satellites of various size and weight, depending on the devices’ spatial and spectral resolutions. Three different domains represent the conventional satellite classes, viz. micro (up to 100 kg), mini (100–500 kg) and large (over 500 kg). The completed analysis clearly shows that it is impossible to carry out all conceivable LEO hyperspectral missions with microsatellites alone. Meanwhile, most of scientific, commercial, and defense hyperspectral tasks can be implemented just with mini-satellites on the basis of conventional technology.

62

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

FIGURE 2.10 Hyperspectral sensors accommodation on satellites of different subclasses as function of imager’s ground resolution and number of bands.

Still, all spatial, spectral, and radiation requirements growing higher, the mass value would exceed the upper limit for mini-satellites, therefore ensuring 200–300 spectral bands with spatial resolution of some meters (especially in the IR region at high-LEO altitudes) would be feasible only with large satellites. Microsatellites would match well only in the visible region, with large limitations on the resolution (the ground resolutions of 30–60 m and the number of spectral bands 30–100, respectively). The spatial and spectral resolutions of the well-known operated hyperspectral satellites EO-1, MightySat II, Proba-1, and TacSat-3 are also drawn in Figure 2.10 with bullet points, their respective weights being put in parentheses. As can be seen, the point’s positions match well with the estimations. However used onboard, the small satellites were special-purpose instruments, with high spectral resolution in a limited number of wavelengths. As an example, the atmospheric ozone mapping by the small satellites in a short spectral range of ozone bands in UV using a simple filter wheel photometer was a successful application of this principle. Similar experiments with poor spatial resolution are available in small satellites [39]. Instrumentation mass and size, affordable by the present technology level, having such an impact on the future mission’s feasibility, big efforts are made to reduce the power consumption, mass, volume, and the unit costs of the hyperspectral sensors under development, as well as to lower demands to the data downlink. Whether these efforts will lead to the emergence of miniature hyperspectral instruments suitable for the small satellites’ missions, only time can show.

2.5 LiDAR AND HYPERSPECTRAL DATA INTEGRATION Virtually all remote sensing techniques, including those presented in the previous paragraphs, rely upon passive sensing of the amount of solar radiation as reflected toward the sensor by clouds, ocean, or solid land, and IR-radiation emitted from the natural or artificial thermal sources on the Earth’s surface. Unlike them, the LiDAR system is an active sensor. Since LiDAR carries a source of radiation of its own, it can determine where and when, in the daytime or at night, to take the measurements. LiDAR is similar to radar in the sense that it also can track any target of interest, from airplanes in flight to thunderstorms. But unlike radar, LiDAR uses short pulses of coherent laser light, with very short wavelengths and high instant capacity, so its laser beam does not disperse while traveling away from the source, as ordinary light does. Besides, lasers offer great advantages over conventional light sources in terms of peak power and narrow spectral bandwidth.

Hyperspectral Sensor Characteristics

63

A primary functional part of LiDAR is the laser, its other components being the scanner, collimator, lens, receiver, amplifiers, samplers, and other optical and electronic elements. The main function of the laser is to generate emission which, having been reflected by the Earth’s surface or the ground objects, returns to the source, to be detected by the highly sensitive receiver. The response time being directly proportional to the target range, it is equivalent to measuring the distance between the emitting source and the reflecting object. The wavelength choice depends on the laser function, as well as on the safety and security requirements. The most commonly used are Nd:YAG lasers, which operate at the wavelengths such as 1550 nm; 1064 nm—near-IR; 532 nm— green light; 355 nm—near-UV. While ground-based LiDAR instruments profile a single viewing site, the airborne LiDAR systems offer one of the most accurate, expedient, and cost-effective ways to capture upward- or downward-looking data over a wider area. Vertical accuracies of less than 5 cm are possible at 500 m above ground, amounting to less than 20 cm at 3-km altitude. This enables LiDAR to be involved in various applications, such as surveys of corridors like pipeline routes, roads, power lines, highways, and railways; urban environments and mapping; flood plain surveys; forestry (mapping of tree canopies); archaeology; seismic exploration; coastal zone surveys; and oil and gas exploration. In the airborne LiDAR systems, the laser range-finder is mounted over an opening in the aircraft floor and scans beneath the aircraft, producing a wide swath over which the distance to the ground is measured. The scanning device controls the direction of the laser pulses’ propagation, to provide coverage of some predetermined scan swath. In most cases, the cross-track scanning is implemented by using the oscillating mirror, while the along-track scanning is by the motion of the carrier aircraft itself along its operational path. All the aircraft movements are recorded by its inertial navigation system (INS), to be used in the data post-processing. The LiDAR can also be installed on a stabilized platform. By merging laser ranging, GPS positioning, and inertial attitude technologies, LiDAR can directly measure the shape of the Earth’s surface beneath the aircraft’s flight path. Data collection rates, accuracy, and other characteristics of some excising and operating airborne LiDARs correspond to the user’s requirements and allow them to carry out various research efforts. A fundamental limitation of airborne laser scanners and imagery consists in their generally straight-down perspective, which makes impossible the accurate and detailed mapping of vertical structures such as cliff faces, coastal ridges, or any vertical side of a natural or man-made structure. To reveal the vertical faces of structures from the air, imaging from the different perspectives should be introduced in the LiDAR data acquisition process by means of some specialized mounting. The airborne applications being limited to comparatively small regions, it is the spaceborne LiDAR systems that can provide continuous geospatial data and offer a truly global view of the Earth’s surface. Furthermore, unlike an airborne laser, a spaceborne one poses no hazard to the general public, even when viewing with the naked eye, binoculars, or small telescopes. The first LiDAR system in space was the LITE (Lidar In-Space Technology Experiment) instrument, designed and built by NASA. From a 260-km Space Shuttle orbit, its pencil-wide laser beam spread to approximately 300 meters wide at the surface—about the size of three football fields. The data obtained by LITE were used in the Earth atmosphere studies, to measure vegetative cover and distinguish various types of surfaces. LiDAR is a very reliable tool for active remote sensing from space. A future spaceborne high-performance imaging LiDAR for global coverage is now under development. While early LiDAR scanning systems were capable of registering a single return pulse for each transmitted pulse, most systems in operation today are able to record multiple discrete returns, which occurs when an emitted LiDAR beam encounters in its path an object like a forest canopy. The data may be classified as first, second, third, or last returns, and bare-Earth. A return might be generated from a reflection caused by the top of the tree, but a sufficient amount of laser light energy is able to reflect from lower portions of the tree, or finally, from the ground. These multiple returns can be analyzed and classified to produce information about the objects as well as the bare ground surface. Buildings, trees, and power lines are individually discernible features and may be

64

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

classified separately. Most often used are the first and last returns, the first one showing the highest features such as the tree canopy, buildings, and so on, and the last generally assumed to refer to the ground level. Being digital, it can be directly processed into a detailed bare-Earth digital elevation model at vertical accuracies within 0.15–1 m. Derived products include contour maps, slope aspect, 3D topographic images, virtual reality visualizations, and more. The LiDAR instruments have an option to measure the intensity of the returned light. This captured information can be displayed as an orthorectified image—a 256-gray level mapping of the light pulse reflectance amplitude. Such an image, comparable with a coarse photograph of the area under survey, can be painted according to the elevation levels. The amplitudes of the laser scan return signals, as measured by LiDAR, do not allow proper reconstruction of the ground reflectance, mainly because of spreading and absorption terms, which depend on the laser-target distance. This prevents not only accurate imaging of the ground, but also the integration of the intensity information with the segmentation/classification process. This difficulty can be overcome by a suitable calibration of the laser scan return values. A number of signal calibration procedures are known, insufficiently precise as they are. However, the radiometric information of the laser scan is relatively poor in terms of information content being the laser light is monochromatic. Some researchers utilize the return intensities to extract more information about the reflecting surfaces than it can be done by using LiDAR solely as a ranging tool. Using the intensity measurements combined with height data for classification of vegetation types, it was found, for example, that the reflectance values, as shown by pure broadleaf forests, are significantly higher than those for pure conifer forests. The exploratory data analysis was carried out to assess the potential of the laser return type and intensity as variables for classification of individual trees or forest stands according to species. The evaluation of the irregular behavior of some major ground indices, fulfilled on the basis of integrating the LiDAR and hyperspectral data, made it possible to refine the technique of discovering new archeological sites. In the land cover classification, it proved effective to process LiDAR intensity data clouds together with the high quality, large-coverage images provided by the airborne cameras. Besides the intensity, the primary products derived from the raw LiDAR data also include the waveform returns. The new full-waveform laser scanning technique offers advanced data analysis options not available in traditional LiDAR sensors. This technique, implemented in the latest generation of scanners, will promote evolution in such areas, as continuous multilayer vegetation modeling; waveform-based feature classification; single-pulse slope determination, and so on. The LiDAR images, including orthophotos, can be seamlessly integrated with MS, hyperspectral, and panchromatic data sets. Combined with Geographic Information System (GIS) data and other surveying information, LiDAR imagery enables the generation of complex geomorphic-structure mapping products, building renderings, advanced 3D Earth modeling, and many more high-quality mapping products. Galileo is one of a few companies providing highly accurate fused image data products. Using high spatial-resolution LiDAR and hyperspectral imaging data, the maps were improved by adding a 3D perspective. As was discovered, the airborne LiDAR can be used for monitoring bio-fluorescence when coupled with a hyperspectral tunable filter [40]. The promising technique of hyperspectral fluorescence imaging, as induced by a 632-nm laser, was applied for assessing the soluble solids content of fruit. The laser-induced hyperspectral bio-fluorescence imaging of different objects is now under study in lab experiments, airborne applications, and fieldwork. The efficiency of the advanced geospatial technology, based on the fusion of the hyperspectral imagery and LiDAR data, was assessed in Canada and the United States, as applied to the mapping of the location and condition of different tree species, for both the effective management of the forested ecosystems and counteraction to invasive tree pests. The hyperspectral imaging was obtained by the AISA Dual system with the Eagle (395–970 nm) and Hawk (970–2503 nm) sensors, enabling simultaneous acquisition in 492 narrow spectral channels. LiDAR data were collected using a TRSI

Hyperspectral Sensor Characteristics

65

Mark II two-return sensor. Also used in the data processing was the collection of ground spectral signatures. German and Israeli scientists are implementing the project on fusion of hyperspectral images and LiDAR data for engineering structure monitoring [41]. They confirm that the assessed degradation of urban materials in artificial structures by exploring possible chemical and physical changes using spectral information across the VIS-NIR-SWIR spectral region (400–2500 nm). This technique provides the ability for easy, rapid, and accurate in situ assessment of many materials on a spatial domain under near real-time condition and high temporal resolution. LiDAR technology, on the other hand, offers precise information about the geometrical properties of the surfaces within the areas under study and can reflect different shapes and formations of the complex urban environment. Generating a monitoring system based on the integrative fusion between hyperspectral sensors and LiDAR data may enlarge the application envelop of each technology separately and contribute valuable information on urban runoff and planning. The LiDAR and hyperspectral data were also investigated from the viewpoint of their potential to predict the canopy chlorophyll and carotenoid concentrations in a spatially complex boreal mixed wood. Using canopy scale application of hyperspectral reflectance and derivative indices, LiDAR data analysis was conducted to identify structural metrics related to chlorophyll concentration. Then the LiDAR metrics and hyperspectral indices were combined to determine whether concentration estimates could be further improved. Integrating mean LiDAR first-return heights for the 25th percentile with the hyperspectral derivative chlorophyll index enabled further strengthening of the relationship to canopy chlorophyll concentration. Maps of the total chlorophyll concentration for the study site revealed distinct spatial patterns, indicative of the spatial distribution of species at the site. The project [42] applied advanced geospatial technology, including high-resolution airborne AISA hyperspectral sensors and LiDAR data collection, in conjunction with analytical applications of GIS, to develop new tools needed for improved species mapping, risk assessment, forest health monitoring, rapid early detection, and management of invasive species. Similar technologies play an increasingly important role in offering accurate, timely, and cost-effective solutions to such problems. Airborne LiDAR has become a fully operational tool for hydrographic surveying. Present-day airborne laser bathymetry systems can simultaneously measure both water depth and adjacent surface topography. hyperspectral imagery from abovementioned Compact Airborne Spectrographic Imager (CASI) (see Section 2.4.2) has proven to be a valuable instrument for coastal measurements and analysis. CASI spectral resolution of 288 bands for each spatial pixel allows for the extraction of a vast amount of information such as water clarity and temperature, bottom type, bathymetry, as well as water quality (chlorophyll, dissolved organic carbon, and suspended minerals), soil types, and plant species [50]. In order to achieve a comprehensive hydrographic capability, the LiDAR and hyperspectral sensors CASI sensors were integrated to provide the different agencies with the relevant information. The Naval Oceanographic Office, for example, uses the airborne laser bathymetry system to collect hydrographic information about the littoral zone for the warfighter and, by adding the hyperspectral capability, enhance its efficiency in rendering a quick and more adequate environmental picture. An interesting research issue pertains to the integration of LiDAR and hyperspectral data to improve the discovery of new archeological sites [43]. Under investigation was the possibility of using such an integrated dataset in evaluation of the irregular behavior of some major ground indices. While hyperspectral data allow identifying specific humidity, vegetation, and thermal conditions in the target area, LiDAR data provide the accurate geometric information. In fact, accurate filtering of the laser scanning data allows the computation of the Digital Terrain Model, while Lambertian-based calibration of LiDAR intensity enhances the automatic data segmentation and thus the detection of possible sites of interest. In order to fully extract hyperspectral information from MIVIS and AISA sensor data, some specific procedures have been implemented together with the adequate computer analysis. Such data processing has been applied to both already discovered archeological sites of the ancient city of Aquileia, and the new areas in its northern part. When integrated, the resulting datasets

66

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

revealed with sensible accuracy the presence of surface/below surface archaeological heritage. As was confirmed by the archaeologists, the present-day available airborne imaging, namely LiDARassisted and the hyperspectral, is the most technologically advanced approach. In some cases, these instruments were put together as an integrated unique sensor mounted on the helicopter. The new remote sensing technologies on a basis of LiDAR data integration with optical hyperspectral and MS imaging data were developed for natural hazard detection and decision support systems. The innovative imagery fusion refines the identification and mapping of past geologic events, such as landslides and faults, while also providing quicker and simpler processes for forecasting and mitigating future environmental hazards. United States researchers used the remote sensing data to improve mapping and characterizing mechanical properties of rocks. In early 2007, simultaneous hyperspectral and LiDAR imagery was  obtained from a plane over Cuprite, Nevada [51]. The imaging instrumentation included the Optech LiDAR, with an operational wavelength of 1.064µm, next-generation hyperspectral sensors—Mapping Reflected-energy Spectrometer, working in the 0.4–2.5 µm spectral range, and the Nikon D2X digital camera to capture high-spatial-resolution true-color images. Results of the data analysis suggested, for some surfaces, a correlation between mineral content and surface roughness, although the LiDAR resolution (∼1 m ground sampling distance) turned out to be too coarse to extract surface texture properties of clay minerals in some of the alluvial fans captured in the imagery. Such experiments may provide valuable information about the mechanical properties of the surface cover in addition to generating another variable of use for material characterization, image classification, and scene segmentation. A short description of the successful research efforts, based on joint hyperspectral sensors and LiDAR data, does not pretend to be complete; it just demonstrates high efficiency of this imaging technique in various applications. Future mission planning should include consideration of determining optimal ground sampling to be used by LiDAR and hyperspectral systems [44,45]. The fusion of LiDAR elevation data and MS and hyperspectral classification results is a valuable tool for imagery analysis and should be explored more extensively.

2.6

SUMMARY AND OUTLOOK

The hyperspectral imaging sensors are designed to provide a detailed analysis of the entire light spectrum from visible light up to far IR. Present-day sensors can break this wavelength range into hundreds, or even thousands, of sections for individual analysis and examine such extremely narrow sections, which makes it possible to look for specific chemical compositions that reflect light only in the chosen wavebands. Hyperspectral sensors simultaneously capture both spectral and spatial information of the scene and provide a fully registered spatial and spectral data cube for every video frame, and are equally well suited for ground, airborne, and space operations. Due to their unique capabilities, they can collect data from hand-held and vehicle-mounted ground systems, low- and high-altitude aircraft, as well as LEO or even GEO satellites, according to the application in question. One shortcoming of flight operations with hyperspectral sensors, also with aerial- and space-based LiDAR, is the need for relatively cloud-free conditions over the target area. To verify the identifications derived from the airborne hyperspectral imagery, ground spectral measurements are often carried out at the sites of interest by hand-held or truck-mounted hyperspectral instruments, providing the ground truth, that is, the material characterization of the natural land samples in UV-VNIR-SWIR-MWIR-LWIR spectral bands. The spaceborne hyperspectral imaging has huge potential with respect to the duration of the observations, Earth’s surface coverage, and abundance of the valuable information. Still there have been few successful hyperspectral space missions due to the complexities mentioned earlier. Described in this chapter were civilian, dual-use, and defense hyperspectral sensors that have been put into orbit, as well as the sensors currently under development or design and some systems planned but never implemented, such as NEMO and Warfighter.

Hyperspectral Sensor Characteristics

67

At present, most of the hyperspectral servicing firms are ready to assist clients in their remote sensing data demands. They can support the entire end-to-end data collection effort, starting with the site identification, observation planning, imaging operations, and data analysis. Typical remote hyperspectral products include spectrally and radiometrically calibrated hyperspectral data cubes; geo-reference files for all data cubes; atmospherically corrected data cubes; field spectra from handheld spectrometers; laboratory spectra of samples; and apparent emissivity data cubes for IR sensors. Without doubt, the hyperspectral data from operating sensors provide an important contribution for the research community [46–48]. The current state-of-the-art hyperspectral data are far ahead of the first tentative steps taken about 30 years ago. New hyperspectral technology, including instruments performance, processing schemes, and calibration, improved significantly. Wellestablished spheres of hyperspectral application include agriculture, forestry, water resources, atmosphere, geology, mineralogy, wetlands, environment, management of coastal waters, defense and security, and urban areas. Hyperspectral mission and sensor performance parameters are derived from the requirements for specific applications. For the spaceborne hyperspectral imaging, along with cost considerations, this leads to the choice of the orbit; re-visit capability; SNR; spatial resolution; data volume per time period; spectral resolution; spectral sampling interval; swath width; radiometric accuracy and stability; spectral range of interest; and required time for the data delivery to the users and processing steps. An individual user’s hyperspectral imagery and related data requirements, which are essential to the event under examination, differ from each other. The most important for all users proved to be not hyperspectral sensors and imagery qualities, but better service and reduced equipment costs. Hyperspectral spaceborne imaging is getting mature and more satellites with hyperspectral sensors are expected to be put into orbit in the next decade. These may include the German EnMAP; Indian TWSat; South African MSMI hyperspectral sensor; satellites with sensors like Hyperon or CHRIS; wide-area synoptic sensors similar to MODIS and MERIS with finer spectral resolution; and NASA HyspIRI and ESA FLEX, in the case of available funding. Future Landsat and NPOESS satellites might have the hyperspectral capability, as well as UK Disaster Monitoring and German RapidEye constellations. Italy, China, Israel, Canada, and other countries have also announced their hyperspectral space projects. Along with this, the development of the airborne and drone hyperspectral imaging for civilian and public-good applications is also anticipated.

REFERENCES 1. Glackin D.L. Peltzer G.R. Civil: Commercial, and international remote sensing systems and geoprocessing, The Aerospace Corporation, AIAA, p. 89, 1999. 2. Kramer H.J. Observation of the Earth and Its Environment, Survey of Missions and Sensors, 4th edition, Springer, Berlin, 2002. 3. Nieke J., Schwarzer H., Neumann A., Zimmermann G. Imaging spaceborne and airborne sensor systems in the beginning of the next century. European Symposium on Aerospace Remote Sensing, IEEE Conference on Sensors Systems and Next Generation Satellites III, 1997. https://ieeexplore.ieee. org/document/1356079/references. 4. Puschell J.J. Hyperspectral imagers for current and future missions, Proceedings of SPIE, 4041, 121–132, 2000. 5. Lillesand T.M., Kiefer R.W., Chipman J.W. Remote Sensing and Image Interpretation, 5th edition, Wiley, NY, p. 763, 2004. 6. Chang C.-I. Hyperspectral Imaging. Technigues for Spectral Detection and Classification, Kluwer Academic/Plenum Publishers, NY, 370, 2003. 7. Borengaser M. Hyperspectral Remote Sensing, Lewis, Boca Raton, FL, 2004. 8. Кучейко А. Российские перспективы в гиперспектре, Новости космонавтики, 222, p. 24, 2001 (Kuchako A., Russian Hyperspectral Perspective, J. Novosti Kosmonavtiki, N 222, p. 24, 2001). 9. Bjorn A. Hyperspectral Chemical and Property Imaging, Patent 20090015686, USPC Class 3482221, 01-15-2009.

68

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

10. Bolton J.F. Full spectral imaging: A revisited approach to remote sensing, SPIE Conference on Remote Sensing, September, Barcelona, Spain, 2003. 11. Bodkin A. Development of a miniature, hyperspectral imaging digital camera, Navy SBIR FY2005.1, Proposal N051-971-0772, 2005. 12. Fletcher-Holmes D.W., Harvey A.R. Real-time imaging with a hyperspectral fovea, Journal of Optics A: Pure and Applied Optics, 7, 298–302, 2005. 13. Skauli T., Kåsen I., Haavardsholm T., Kavara A., Tarabalka Y., Farsund O. Status of the Norwegian hyperspectral technology demonstrator, NATO OTAN, RTO-MP-SET-130, Norwegian defence research establishment. 14. Yang C., Everitt J.H., Davis M.R., Mao, C. A CCD camera-based hyperspectral imaging system for stationary and airborne applications, Geocarto International Journal, 18(2), 71–80, 2003. 15. Anger C.D. Airborne hyperspectral remote sensing in the future, Proceedings of the 4th International Airborne Remote Sensing Conference and Exhibition/21st Canadian Symposium on Remote Sensing, Vol. 1, ERIM International Inc., Ann Arbor, Michigan, USA, 1–5, 1999. 16. Buckingham R., Staenz K., Hollinger A. A review of Canadian airborne and space activities in hyperspectral remote sensing, Canadian Aeronautics and Space Journal, 48(1), 115–121, 2002. 17. Jianyu W., Rong S., Yongqi X. The development of Chinese hyperspectral remote sensing technology, SPIE, 5640, 358, 2005. 18. Briottet X. et al. Military applications of hyperspectral imagery, Proceedings of SPIE Defense & Security Symposium, paper No. 62390B, 2004. 19. Jason S., Cutter M., Meerman M., Curiel A.S. Low cost hyperspectral imaging from a microsatellite, 15-th Annual AIAA/USU Conference on Small Satellites, SSC01-II-1, Utah, 2001. 20. Curiel A.S., Cawthorne A., Sweeting M. Progress in small satellite technology for earth observation missions, 5-th IAA Symposium on Small Satellites for Earth Observation, IAA-B5-0301, Berlin, 2005. 21. Cutter M. Review of a small satellite hyper-spectral mission, 19-th Annual AIAA/USU Conference on Small Satellites, SSC05-IV-2, Utah, 2005. 22. Ortenberg F., Guelman M. Small satellite’s role in future hyperspectral earth observation missions, Acta Astronautica, 64(11–12), 1251–1262, 2009. 23. Pearlman J., Barry P., Segal C., Shepanski J., Beiso D., Carman S. Hyperon, a space-based imaging spectrometer, IEEE Transactions on Geoscience and Remote Sensing, 41(6), 1160–1173, 2003. 24. Vidi R., Chiaratini L., Bini A. Hyperresolution: An hyperspectral and high resolution imager for Earth observation, Proceedings of the 5th International Conference on Space Optics (ICSO 2994), Toulouse, France, 105–111, 2004. 25. Davis C. et al. Ocean PHILLS hyperspectral imager: Design, characterization and calibration, Optics Express, 10(4), 210–221, 2002. 26. Hollinder A. et al. Resent developments in hyperspectral environment and resource observer (HERO) mission, Proceedings of the International Geoscience and Remote Sensing Symposium and 27th Canadian Symposium on Remote Sensing, July, Denver, Colo, 1620–1623, 2006. 27. Pretil G., Cisbanil A., De Cosmo V., Galeazzi C., Labate D., Melozzi M. Hyperspectral instruments for earth observation, International Conference on Space Optics (ICSO), Toulouse, France, 2008. 28. Cutter M., Sweeting M. The CHRIS hyperspectral mission—Five years in-orbit experience, IAC07-B1.3.02, 58th International Astronautical Congress, September 24–28, Hyderabad, India, 2007. 29. Stuffer T. et al. The EnMAP hyperspectral imager—An advanced optical payload for future applications in earth observation programmes, Acta Astronautica, 61, 115–120, 2007. 30. Sigernes F. et  al. Proposal for a new hyper spectral imaging micro satellite: SVALBIRD, 5-th IAA Symposium on Small Satellites for Earth Observation, IAA-B5-0503, Berlin, 2005. 31. Morea G.D., Sabatini P. Perspectives and advanced projects for small satellite missions at Carlo Gavazzi Space, Proceedings of the 4S Symposium Small Satellites Systems and Services, La Rochelle, France, 2004. 32. Poinsignon V., Duthil P., Poilve H. A superspectral micro satellite system for GMES land cover applications, Proceedings of the 4S Symposium Small Satellites Systems and Services, La Rochelle, France, 2004. 33. Haring R.E., Pollock R., Cross R.M., Greenlee T. Wide-field-of-view imaging spectrometer (WFIS): From a laboratory demonstration to a fully functional engineering model, Proceedings of SPIE, 4486, 403–410, 2002. 34. Schoonwinkel A., Burger H., Mostert S. Integrate hyperspectral, multispectral and video imager for microsatellites, 19-th Annual AIAA/USU Conference on Small Satellites, SSC05-IX-6, Utah, 2005.

Hyperspectral Sensor Characteristics

69

35. Cooley T., Lockwood R., Gardner J., Nadile R., Payton A. ARTEMIS: A rapid response hyperspectral imaging payload, Paper No. RS4-2006-5002, 4th Responsive Space Conference, April 24–27, Los Angeles, CA, 2006. 36. Li J., Sun F., Schmit T., Venzel W., Gurka J. Study of the hyperspectral environmental suit (HES) on GOES-R, Proceedings of the20th International Conference on Interactive Information and Processing System forMeteorolgy, Oceanography, and Hydrology, January, Seattle, Wash., Paper p2.21.6, p. 6, 2004. 37. Astapenko V.M., Ivanov V.I., Khorolsky P.P. Review of current status and prospects of hyperspectral satellite imaging, Space Science and Technology, Kyiv, Ukraine, 8(4), 73, 2002. 38. Davis T., Straight S. Development of the tactical satellite 3 for responsive space missions, Proceedings of the 4th Responsive Space Conference, April 24–27, Los Angeles, CA, p. 10, 2006. 39. Guelman M., Ortenberg F., Shiryaev A., Waler R. Gurwin-Techsat: Still alive and operational after nine years in orbit, Acta Astronautica, 65(1–2), 157–164, 2009. 40. Liu M., Zhang L., Guo E. Hyperspectral laser-induced fluorescence imaging for nondestructive assessing soluble solids content of orange, Computer and Computing Technologies in Agriculture, I, 51–59, 2008. 41. Brook A., Ben-Dor E., Richter R. Fusion of hyperspectral images and LIDAR data for civil engineering structure monitoring, Commission VI, WG VI/4, Proceedings ISPRS XXXVIII-1-4-7, W5, paper 127, 2009. 42. Souci J., Hanou I., Puchalski D. High-resolution remote sensing image analysis for early detection and response planning for Emerald Ash Borer, Photogrammetric Engineering and Remote Sensing, August, 905–907, 2009. 43. Johnson J.K. Sensor and data fusion technologies in archaeology, ISPRS Workshop on Remote Sensing Methods for Change Detection and Process Modeling, Germany, 2010. 44. Fujii T., Fukuchi T. Laser Remote Sensing, Taylor & Francis, Abingdon, UK, p. 888, 2005. 45. Zhang J. Multi-source remote sensing data fusion: Status and trends, International Journal of Image and Data Fusion, 1(1), 5–24, March 2010. 46. Nieke J., Seiler B., Itten K., Ils Reusen I., Adriaensen S. Evaluation of user-oriented attractiveness of imaging spectroscopy data using the Value-Benefit Analysis (VBA), 5th EARSeL Workshop on Imaging Spectroscopy, April 23–25, Bruges, Belgium, 2007. 47. Puschell J. Hyperspectral imagers for current and future missions, Proceedings of SPIE, 4041, 121–132, 2000. 48. Buckingham R., Staenz K. Review of current and planned civilian space hyperspectral sensors for EO, Canadian Journal of Remote Sensing, 34(Suppl. 1), S187–S197, 2008. 49. Schulenburg N., Riley D., Kupferman P., Bryant G. On using SEBASS as a Mineral Exploration Tool, GRSG AGM—Exploration Remote Sensing, Geologic Society of America, 2009. https://www.grsg.org. uk/abstracts/on-using-sebass-as-a-mineral-exploration-tool/. 50. Lopez R., Frohn R. Remote Sensing for Landscape Ecology: New Metric Indicators—Monitoring, Modeling, and Assessment of Ecosystems, Taylor & Francis-CRC Press, Boca Raton, FL, 2017. 51. Riley D., Hecker C. Mineral mapping with airborne hyperspectral thermal infrared remote sensing at Cuprite, Nevada, USA, Chapter 24, In: Thermal Infrared Remote Sensing, Springer, 2013.

3

Hyperspectral Remote Sensing in Global Change Studies Jiaguo Qi, Yoshio Inoue, and Narumon Wiangwang

CONTENTS 3.1 3.2

Introduction ............................................................................................................................ 72 Hyperspectral Sensors and Characteristics ............................................................................ 72 3.2.1 Spaceborne Systems ................................................................................................... 72 3.2.2 Airborne Systems ....................................................................................................... 75 3.2.3 Ground-Based Systems ............................................................................................... 75 3.3 Hyperspectral Remote Sensing Methods................................................................................ 76 3.3.1 Support Vector Machines ........................................................................................... 76 3.3.2 Kernel Fisher Discriminant Analysis ......................................................................... 77 3.3.3 Matched Filtering ....................................................................................................... 77 3.3.4 Libraries Matching Techniques .................................................................................. 77 3.3.5 Derivative Spectroscopy ............................................................................................. 78 3.3.6 Narrow Band Spectral Indices.................................................................................... 78 3.3.6.1 Normalized Difference Vegetation Index: NDVI ........................................ 78 3.3.6.2 Yellowness Index: YI ................................................................................... 79 3.3.6.3 Normalized Difference Water Index: NDWI............................................... 79 3.3.6.4 Red-Edge Position Determination: REP...................................................... 79 3.3.6.5 Crop Chlorophyll Content Prediction .......................................................... 79 3.3.7 Neural Network .......................................................................................................... 79 3.4 Global Change Requirements and Applications .....................................................................80 3.4.1 Global Change Requirements .....................................................................................80 3.4.2 Global Change Applications .......................................................................................80 3.4.2.1 Water Quantity and Quality .........................................................................80 3.4.2.2 Carbon Sequestration and Fluxes ................................................................ 81 3.4.2.3 Greenhouse Gas Emissions.......................................................................... 82 3.4.2.4 Atmospheric Chemistry ............................................................................... 82 3.4.2.5 Vegetation Ecology ...................................................................................... 83 3.4.2.6 Vegetation Biochemical Properties .............................................................. 83 3.4.2.7 Invasive Plant Species Detection ................................................................. 83 3.4.2.8 Vegetation Health.........................................................................................84 3.5 Hyperspectral Remote Sensing Challenges ............................................................................84 3.5.1 System Design Challenges..........................................................................................84 3.5.2 Processing and Visualization Challenges ................................................................... 85 3.5.3 Data Volumes and Redundancy .................................................................................. 85 3.5.4 Radiometric Calibration ............................................................................................. 86 3.5.5 Methodological Challenges ........................................................................................ 87 3.6 Discussion and Future Directions ........................................................................................... 87 References ........................................................................................................................................ 87

71

72

3.1

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

INTRODUCTION

A few decades ago, hyperspectral imagery data and processing software were available to only spectral remote sensing experts. Nowadays, one of the fastest growing technologies in the field of remote sensing has been investing on the research and development of hyperspectral sensors for data acquisition and software for data analysis [1]. Unlike multispectral imaging systems (e.g., Landsat, SPOT, IKONOS) that capture reflected or emitted incoming radiation from the Earth’s surface in a few broad wavelength bands across the electromagnetic spectrum, hyperspectral imagers measure reflected radiation at numerous narrow, contiguous wavelength channels. The substantially finer spectral resolution data from hyperspectral sensors enhance the capability to characterize the Earth’s surface more effectively than do the broadband multispectral data [2]. The distinction between hyperspectral and multispectral is based on the narrowness and contiguous nature of the measurements, not the “number of bands” [3]. For example, a sensor that measures only 20 spectral bands can be considered hyperspectral if those bands are narrow (e.g., 10 nm) and contiguous where there is useful content to be collected. On the other hand, if a sensor measures 20 wider spectral bands (e.g., 100 nm), or is separated by non-measured wavelength ranges, the sensor is no longer considered hyperspectral [1]. The detailed contiguous range of spectral bands of a hyperspectral sensor provides an ability to produce a contiguous “spectrum,” which is one of the characteristics that distinguishes it from multispectral sensors. Radiances measured by multispectral sensors are generally adequate for rough discrimination of surface cover into categories; however, they are rather limited in the amount of quantitative information that can be inferred from the spectral content of the data. The spectra, or spectral reflectance curves, from hyperspectral remote sensors provide much more detailed information about absorption regions of the surface of interest, very much like the spectra that would be measured in a spectroscopy laboratory. This unique characteristic of hyperspectral data is useful for a wide range of applications, such as mining, geology, forestry, agriculture, and environmental assessment. This chapter is to focus on existing hyperspectral remote sensing systems, global change requirements, application examples, and challenges ahead.

3.2 HYPERSPECTRAL SENSORS AND CHARACTERISTICS In the 1970s, space-based multispectral remote sensors were launched and produced images of the Earth’s surface. Even with only a few broad wavelength bands, the images greatly improved our understanding of our planet’s surface. The idea of developing hyperspectral, imaging sensors, also known as imaging spectroscopy, emerged in the early 1980s to improve our ability to better characterize the Earth’s surface. The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) developed at the NASA Jet Propulsion Laboratory in California was the first spectrometer being used on moving platforms such as aircraft. In the early imaging spectroscopy era, most of the hyperspectral sensors were mounted on aircraft (e.g., AVIRIS). After decades of research and development, hyperspectral technology was expanded to space-based remote sensing systems and several satellite hyperspectral sensors were proposed and subsequently launched. The very first spaceborne hyperspectral sensors were NASA’s Hyperion sensor on the Earth Observing-1 (EO-1) satellite, and the U.S. Air Force Research Lab’s Fourier Transform Hyperspectral Imager (FTHSI) on the MightySat II satellite. With more satellite-based sensors being planned, more hyperspectral imagery will be available to provide near global coverage at regular repeated cycles [1] suitable for global change studies.

3.2.1

sPaceborne systems

Spaceborne hyperspectral sensors aboard satellites may provide continuous acquisition of the Earth’s surface images at lower relative cost. However, wide spatial coverage by a hyperspectral sensor is

Hyperspectral Remote Sensing in Global Change Studies

73

often compromised with its spatial resolution or ground sampling interval and other challenges. Consequently, repeated hyperspectral images are not widely available. Table 3.1 lists currently available spaceborne hyperspectral instruments, with a wide range in the number of spectral bands, spectral range, and swath width. The FTHSI program initiated in 1995, was successfully launched in July 2000, maybe the only Department of Defense (DoD) space-based hyperspectral imager to discern spectrally unique objects with the Fourier transform technique. The width of the image footprint was 13 km with a spatial resolution 30 m, covering a spectral range from 450 to 1050 nm by 256 spectral bands (Table 3.1). In November 2000, NASA launched the Hyperion sensor onboard the EO-1 satellite as part of a 1-year technology validation/demonstration mission. The Hyperion imaging spectrometer had a 30 m spatial resolution, 7.7 km swath width, and 10 nm contiguous spectral resolution (Table 3.1). With its high radiometric accuracy of 220 spectral bands, complex landscapes of the Earth’s surface could be imaged and spectrally characterized. The Compact High Resolution Imaging Spectrometer (CHRIS) was launched in October 2001 on board the PROBA platform (Table 3.1). The sensor was developed by in the United Kingdom with support from the British National Space Centre. The sensor was designed for the study of atmospheric aerosols, land surfaces, and coastal and inland waters with its 62 spectral bands ranging from 400 to 1050 nm and a spatial resolution of 17 m. Despite the fact that the mission was designed for a 1-year life span, the sensor was in operation for more than ten years [4]. The Global Imager (GLI) was part of the ADEOS II mission, an international satellite mission led by the Japanese Aerospace Exploration Agency (JAXA) with participations from the United States (NASA) and France (CNES—Centre Nationale d’Etudes Spatiales; Table 3.1). Its spectral range was from 250 to 1000 nm and its image size ws very large (1600 km). The GLI mission was to collect data to aid better understandings of water, energy, and carbon circulations in order to contribute to global environmental change studies. ADEOS II was launched on December 14, 2002, but unfortunately, the mission ended 10 months later, due to a failure of the solar panel on October 24, 2003. The Ozone Monitoring Instrument (OMI) flown on the EOS Aura spacecraft was designed to measure atmospheric composition (Table 3.1). The sensor was launched on July 15, 2004, into an ascending node 705 km sun-synchronous polar orbit. The OMI was a nadir pushbroom hyperspectral imaging sensor that observed solar backscatter radiation in the UV and visible wavelengths (264– 504 nm). It had 780 spectral bands with a swath large enough to provide global coverage in one day (14 orbits) at a spatial resolution of 13 × 24 km at nadir. The key air quality components include NO2, SO2, and aerosol characteristics, as well as ozone profiles. The Hyperspectral Imager for the Coastal Ocean (HICO), the Navy’s “Sea Strike” was designed and built by the Naval Research Laboratory (NRL) to be the first spaceborne imaging spectrometer optimized for scientific investigation of the coastal ocean and nearby land regions with high signalto-noise ratios in the blue spectral region and full coverage of water-penetrating wavelengths (Table 3.1). Due to the fact that water absorbs most of the light in the electromagnetic spectrum, visible light is the only part of the spectrum that sufficiently penetrates the water column to sense the water and the bottom surface properties. The sensor was launched on the H-2 Transfer Vehicle (HTV) and was rendezvoused with the International Space Station (ISS) in September 2009. The sensor has been serving as a spaceborne hyperspectral method to detect submerged objects, to provide environmental data products to Naval forces, and to develop the coupled physical and bio-optical models of coastal ocean regions globally. A few more sensors have been proposed for the near future launches, including PRecursore IperSpettrale della Missione Applicativa (PRISMA) under development by the Italian Space Agency (ASI). Listed in Table 3.1, PRISMA is a pushbroom hyperspectral sensor and is optimized to derive information about land cover, soil moisture and agricultural land uses, quality of inland waters, status of coastal zones, pollution, and the carbon cycle [5]. The hyperspectral instrument is to acquire images in 250 spectral bands at 30 m spatial resolution. When combined with a panchromatic camera, a higher spatial resolution (5 m) could be produced.

HypSEO (MITA) MSMI (SUNSAT) PRISMA Global Imager (ADEOS-2) WFIS

EnMAP (Scheduled 2014)

∼390 605 523 470 470 650 650 450 660 695 803 705

705 575 580 705

Hyperion (EO-1) FTHSI (MightySat II) CHRIS (PROBA) OMI (AURA) HICO

COIS (NEMO) HIS (SIMSA) Warfighter-1 (OrbView-4)

Altitude (km)

Instrument (Satellite)

30 25 8 8 30 30 20 15 30 250–1000 1400

30 30 25 13,000 92

Pixel Size (m)

210 220 200 80 94 155 210 200 250 36 630

220 256 19 780 102

Number of Bands

TABLE 3.1 Operational and Planned Satellite Hyperspectral Instruments

400–2500 430–2400 450–2400 3000–5000 420–1000 900–2450 400–2500 400–2350 400–2500 380–1195 360–1000

400–2500 450–1050 400–1050 270–500 380–900

Spectral Range (nm)

10 20 11 25 5–10 10–20 10 10 10 10–1000 1–5

10 10–50 1.25–11.0 0.45–1.0 5.7

Spectral Resolution (nm)